Rseg

A review on integrated active steering and braking control for vehicle yaw stability system.
pdf
A review on the platform design, dynamic modeling and control of hybrid UAVs.pdf
Electric vehicle fleet management in smart grids A review of services.pdf
Integrated Vehicle Dynamics Control.pdf
Review of course keeping control system for unmanned surface vehicle.pdf
Review of Dynamic Research for NVH Problems Related to Automotive Driveline.pdf
Review of research on agricultural vehicle autonomous guidance.pdf
Review of the State of Development of Advanced Vehicle Control Systems AVCS (1).pdf
Special issue on the 4th International Tyre Colloquium.pdf
State of the art survey active and semi-active.pdf
The vehicle routing problem A taxonomic review.pdf
Tire Model Application and Parameter Identification-A Literature Review.pdf
Validation of vehicle dynamics simulation models a review.pdf
Various Approaches for Driver and Driving Behavior Monitoring.pdf
A review of active yaw control system for vehicle handling and stability enhancement.pdf
A review of design and control of automated guided.pdf
A review of experimental techniques for NVH analysis on a.pdf
A review of fuzzy methods in automotive.pdf
A review of human sensory dynamics for application to models.pdf
A review of system dynamics models applied in transportation.pdf
Jurnal
Teknologi
Full paper
Full paper
Jurnal
A Review
Teknologi
on Integrated Active Steering and Braking Control for Vehicle Yaw

Stability System
M.K. Aripina,*, Y.M. Samb, A. D. Kumeresanb, M.F. Ismailc, Peng Kemaod
aControl,
Instrumentation & Automation Dept., Faculty of Electrical Engineering, UniversitiTeknikal Malaysia Melaka, Melaka
of Control and Mechatronics, Faculty of Electrical Engineering, Universiti Teknologi Malaysia, 81310 UTM Johor Bahru, Johor Malaysia
cIndustrial Automation Section, Universiti Kuala Lumpur Malaysia France Institute
d
Temasek Laboratory, National University of Singapore
bDepartment
*Corresponding author:khairiaripin@utem.edu.my,khairi32@live.utm.my
Article history
Abstract
Received 1 January 2014

Received in revised form:
17 August 2014
Accepted 24 August 2014
A review study on integrated active steering and braking control for vehicle yaw stability system
is conducted and its finding is discussed in this paper. For road-vehicle dynamic, lateral dynamic
control is important in order to determine the vehicle stability. The aw stability control system is
the prominent approach for vehicle lateral dynamics where the actual yaw rate and sideslip should
be tracked by the controller close to the desired response. To improve the performance of yaw
stability control during steady state and critical driving conditions, a current approach using active
control of integrated steering and braking could be implemented. This review study discusses the
vehicle models, control objectives, control problems and propose control strategies for vehicle
yaw stability control system. In the view of control system engineering, the transient performances
of tracking control are essential. Based on the review, this paper discusses a basic concept of
control strategy based on the composite nonlinear feedback (CNF) and sliding mode control
(SMC) whichcan be proposed for integrated active steering and braking control in order to
improve the transient performances of the yaw rate and sideslip tracking control in the presence of
uncertainties and disturbances.
Graphical abstract
Fx1
d
Fx2
Fy2
Fy1
vx
lf
r Mz
vy
Fx3
Fy3
Fx4
lr
Fy4
Keywords: Vehicle yaw stability; active steering control, active braking control, lateral dynamics
control
2014Penerbit UTM Press. All rights reserved.

.
1.0 INTRODUCTION
This paper presented a review study on integrated active steering
and braking control for yaw stability control system of roadvehicles. The main objective of this paper is to provide an
overview and comprehensive understanding of the main elements
in yaw stability control system and its control strategies in
improving vehicle handling and stability performances.
In vehicle dynamics studies, vehicle stability is determined
by the lateral dynamic motion where yaw stability control system
is one of the prominent approaches for lateral dynamics control
and it can be controlled by using steering and braking subsystem.
In yaw stability control strategy, it is required that designed
controller could control the actual vehicle yaw rate and sideslip
angle with fast responses and good tracking capability in
following the desired responses. Improper commands to steering
and braking subsystems by the driver during critical driving
manoeuvres can cause the vehicle to become unstable and could
lead to accidents. Therefore, an active control is essential to assist
the driver to keep the vehicle stable on the desired path. By

keeping the actual yaw rate and sideslip close to the desired
responses, the active steering and braking control could improve
the handling and stability performances. However, in real
environment, the dynamics of the vehicle are incorporated with
disturbances and uncertainties due to varying parameters which
can influence the yaw stability control performances.
This review study is focused on the main elements of
integrated active steering and braking control and its control
strategies for particular problems. The reviews begin with the
vehicle dynamics models,presented in Section 2. The yaw
stability control objectives and problems are reviewed in Section
3 and followed by control strategies of integrated active steering
and braking control system in Section 4. Section 5 discussed the
basic concept of high performance robust tracking control strategy
and lastly the paper ends with the conclusion in Section 6.
71:2 (2014) 105111 | www.jurnalteknologi.utm.my | eISSN 21803722 |
106
M.K Aripin et al. / Jurnal Teknologi (Sciences & Engineering) 71:2 (2014) 105-111
2.0 VEHICLE MODELS

In a control system design requirement, the modeling of a
dynamic system is vital as it represents the behaviour and
characteristics of the system being controlled. In vehicle dynamics
studies, the mathematical modeling of the vehicle dynamic
motions is modeled based on physical laws that describe the
forces and moments acting on the vehicle body and tires. This
section will brief the dynamics of nonlinear 7 degree-of-freedom
(DOF) vehicle model, linear bicycle model, and a reference model
with the main assumption that the vehicle is moving with constant
speed on plane surface where the roll and pitch motionsare
neglected.
I w i Rwi Fxi Tei Tbi
where i is wheel angular acceleration, Rwi is wheel radius, I wi

is wheel inertia, Tbi is braking torque and Tei is driving torque.
The nonlinear lateral and longitudinal tire forces, Fyi and Fxi can
be described using the nonlinear tire model.Pacejka tire model is
very prominent and widely used in representing nonlinear tire
forces for full vehicle model as well as CarSim vehicle model as
utilized in [1, 2, 4-8, 11, 12, 14, 18, 20-24]. The details of Pacejka
tire model can be referred to in [25].
2.1 Nonlinear 7 DOF Vehicle Model
Fx1
Vehicle motion with nonlinear tire forces represents a nonlinear

system. Nonlinear full vehicle model is usually used to simulate a
real vehicle in order to validate the controller that has been
designed. It could have different number of DOF where
itrepresents the complex vehicle according to their control
structure and algorithm. As utilized in [1-5] the 7 DOF dynamic
vehicle model would record thelongitudinal,lateral and yaw
dynamic motions of the vehicle body while the four tires
represented four wheel dynamic motions.
Another non-linear full vehicle model that wasextensively
used for integrated active steering and braking control is the 8
DOF vehicle model as discussed in [6-12]. In this model, the
dynamics of the 7 DOF vehicle model is extended with roll
motion included. Besides that, the nonlinear vehicle model
possesses 27 DOF based on commercial vehicle dynamics
software such as CarSim, is also utilized in order to evaluate the
proposed controller as implemented in [13-18]. However, in
vehicle yaw stability system for vehicle handling and stability
improvement, the nonlinear 7 DOF vehicle model that neglected
roll and pitch motion is sufficient for simulation and evaluation of
the controller design. The typical full vehicle model in cornering
manoeuvre is shown in Figure 1 while the vehicle parameters are
described in Table 1. On the plane surface (planar motion) where
the roll and pitch motions are neglected, the dynamics equations
for 3 DOF of longitudinal, lateral and yaw motions of the vehicle
body are described based on Newtons 2nd law of rigid body as
follows [19];
Longitudinal motion:
max m(v x rv y ) ( Fx1 Fx 2 ) cos f
Fx3 Fx 4 ( Fy1 Fy 2 ) sin f

Lateral motion:
ma y m(v y rvx ) ( Fx1 Fx 2 ) sin f
( F y1 F y 2 ) cos f F y 3 F y 4
Yaw motion:
I z r l f ( F y1 cos f F y 2 cos f Fx1 sin f
Fx 2 sin f ) l r ( F y 3 F y 4 ) M z
(2)
d Fx1 cos f Fx 2 cos f
2 Fy1 sin f Fy 2 sin f Fx3 Fx 4
Fx2
Fy2
vx
lf
r Mz
vy
Fx3
Fy3
Fx4
lr
Fy4
d
Figure 1 Full vehicle model
Table 1 Vehicle parameters

symbol
m
lf
lr
d
Iz
Cf
Cr
v
Mz
description
vehicle mass
distance from front axle to centre of gravity
distance from rear axle to centre of gravity
width track
moment of inertia
front tire cornering stiffness
rear tire cornering stiffness
vehicle speed
yaw moment
front wheel steer angle
r
vx
vy
Fxi
Fyi
vehicle sideslip
yaw rate
longitudinal velocity
lateral velocity
longitudinal tire forces
lateral tire forces
2.2 Linear Bicycle Model

(3)
where
Mz
Fy1
(1)
(5)
(4)
In vehicle dynamics studies, each wheel represents 1 DOF. Thus,

there are 4 DOF for road vehicle with 4 wheels. The dynamic
motion for each wheel is described as follows [19];
The classical linear bicycle model as shown in Figure 2 is the

linearized model from the nonlinear full vehicle model. It is
prominently used to analyse and design the yaw stability
controller for vehicle lateral dynamics as mentioned in [5-7, 9-11,
16, 18, 22, 26-28]. Based on assumptions as discussed in the
literature, linear bicycle model consists of 2 DOF for lateral and
yaw motion only which can be represented in linear state space
model as follows [19];
107
(6)
a12 a 21 a11 a 22
a12 (a 21e1 a11e2 )
(10)
tr
a12 a 21 a11 a 22
.k r
a 22 (a 21e1 a11e2 )
(11)
k 0
kr
(12)
v
l (1 k us v 2 )
(13)
and k us is understeer parameter as given in equation (8). Notice

where the parameters v, C f , C r , m, l f , l r and I z are the same as
also that a11 , a12 , a 21 and a 22 in equation (10) and (11) are as
described in equation (6).
shown in Table 1. The input of the model is the front wheel steer
angle f while the outputs or controlled variables are vehicle
3.0 YAW CONTROLOBJECTIVES& PROBLEMS
sideslip angle and yaw rate r . Notice that vehicle speed v is

always assumed constant which means the vehicle is not involved
with accelerating and braking.
y
Fyr
v
r
Fyf
Mz
lr
lf
Figure 2 Bicycle model
2.3 Reference Model

The reference model is essential in order to generate the desired
response of vehicle side slip and yaw rate r so that the
designed controller is able to track these desired responses.
According to [19] a desired model to be followed by an actual
vehicle model is based on linear bicycle model in steady state
condition. Since the vehicle sideslip is the deviation angle
between vehicle longitudinal axis and its motion direction, the
desired sideslip d for steady state condition is always zero. The
desired yaw rate response, rd in steady state cornering can be
approximated as first order system which is obtained in term of
steering angle f , vehicle speed v and vehicle parameters as
follows [19];
rd
v
(l f l r ) k us v 2
. f
(7)
where k us is known as understeer parameter and defined as

follows [19];
m(l r C r l f C f )
k us
(8)
(l f l r )C f C r
Hence, the state equations of the desired vehicle model can be
expressed as follows [3, 8, 13, 29,30];
x Ad x d E d f
d t

rd 0
where
0

d t
1 r
k
d r
tr
tr
(9)
In vehicle yaw stability control system, the variables of yaw rate r

and sideslip are very important to determine the stability of the
vehicle. As discussed in [31], control objectives of yaw stability
control may be classified into three categories i.e. yaw rate
control, sideslip control or combined yaw rate & sideslip control.
Yaw rate control is more concerned with improving the
manoeuverability or handling of vehicle while sideslip control
will determine the lateral stability. In order to improve the vehicle
handling and stability performances, a combined yaw rate and
sideslip tracking control are essential.
In integrated active steering and braking control, steering angle
f and yaw moment M z are considered as two independent
inputs for steering and braking actuators[32]. The desired yaw
rate and sideslip that were generated by reference model should
be able to be tracked by the yaw stability controller. Based on the
tracking error, the yaw stability controller will determine an
appropriate correction of steering angle c and yaw moment
M z as implemented in [2, 3, 5, 10, 11, 16, 17, 27, 29, 30].

In integrated active steering and braking control for vehicle
yaw stability, one of the main problems that influence yaw rate
and sideslip tracking control are uncertainties causedby variations
of dynamics parameters such as vehicle speed [1, 6, 22], road
surface adhesion coefficients [4], tire cornering stiffness [9, 10,
14, 22, 29], vehicle mass and moment of inertia [14]. Besides that,
external disturbance such as lateral crosswind [8, 17] may affect
the trackingcontrol of the desired yaw rate and sideslip. Therefore,
the uncertainties of dynamic parameters and crosswind
disturbance that have great influencesto the vehicle handling and
stability performances must be overcome. In order toachieve yaw
stability controlobjectives in the presenceof these problems, an
appropriate control strategy is essential.
4.0
INTEGRATED ACTIVE STEERING & BRAKING

CONTROL STRATEGIES
In vehicle yaw stability system, yaw rate and sideslip tracking

control can be achieved via active chassis control i.e. active
steering, active braking or integrated active steering and braking
control. In this section, the control structure and algorithms of
integrated active steering and braking control that was designedis
reviewed and discussed.
4.1Control Structure
108
In recent years, an integration of active front steering (AFS) and

braking that are based on direct yaw moment control (DYC)
method has become a popular research topic in vehicle dynamics
studies especially for lateral and yaw stability control as reported
in [1, 5, 9, 10, 20-22, 26, 28-30, 33]. As discussed earlier, two
independent inputs to the vehicle i.e. corrective steer angle and
yaw moment as illustrated in Figure 3 will give advantages for
vehicle handling and stability performances improvement during
steady state and critical driving manoeuvres.
fd c
fd
desired
vehicle model
d , rd
e1 d
e2 r rd
c
controller
M z
,r
actual vehicle
model
Figure 3Control structure of integrated AFS-DYC control
To overcome the presence of uncertainties and crosswind

disturbances as discussed above, robust algorithms for the
controller is a necessity.
4.2Yaw Stability Control Algorithms
In order to implement robust yaw stability controller, a linear state
space model of integrated active front steering and direct yaw
moment control as described in equation (14) is utilized for the
controller analysis. Based on this model, control algorithm is
designed to cater to the uncertainties and external disturbances.
Cr lr C f l f
C f Cr
1

mv
mv2
2
C l C l
C f l f Cr lr2
f f
r r r
Iz
I zv
Cf
Cf
0
c mv
mv
fd
C l
f f 1 M z C f l f
Iz
Iz
I z
(14)
Studies in [1, 4,22] considered the road adhesion coefficient,

vehicle speed, cornering and braking stiffness as uncertainties
parameters. Gain-scheduled algorithm is implemented in [1]
whilestate feedback robust controller is designed based on linear
matrix inequality (LMI) in [4, 22] to ensure the robustness of
performance for integrated active front steering and active
differential braking control. In integrated vehicle dynamics
control of [6], phase plane method was used for dynamic stability
control while sliding mode control algorithm was utilized for the
active front steering in order to cater to the uncertainties. Due to
the high nonlinearity and uncertainty of the wheel dynamics, the
sliding mode control algorithm was utilised in [7] for active
braking and active front steering control. Furthermore, it works
with MR suspension system for ride comfort and vehicle stability
improvement.
As discussed in [9], cornering stiffness are considered as
uncertainty parameter. To ensure the robustness of the designed
controller, an adaptive integrated control algorithm based on
direct Lyapunov method is designed for integrated active front
steering and direct yaw moment control. In [10] an optimal
guaranteed cost control (OGCC) technique is utilized for active
front steering and direct yaw moment control to cater for the
uncertainty of tire cornering stiffness during variations of driving

conditions. In [14] an integrated control of active differential and
active front steering had utilized PI control scheme based on yaw
rate error. Tire cornering stiffness, vehicle mass and moment of
inertia are treated as uncertainties parameters.
Integrated control schemes that coordinated the steering,
braking and stabiliser may consist of main loop and servo loop
[15]. The main loop controller computes the stabilizing
forces/moments while servo loop controller consistsof the
forces/moments distribution. In this study, sliding mode controller
is utilized for stabilising the forces and moments to ensure the
robustness of controller. In unified chassis control scheme [16],
the tire cornering stiffness is considered as uncertainty parameter
and sliding mode control algorithm is used to design robust direct
yaw moment control in order to improve lateral stability and
vehicle handling. For global chassis control [20]a robust
controller is designed based on nonlinear adaptive H control
theory to attenuate an external disturbances and uncertainties of
vehicle mass, moment of inertia and centre of gravity position.
Similarly in [6], the study in [27] used sliding mode control
for active front steering in integrated control with electronic
stability program to ensure robustness against uncertainties of
vehicle mass, speed and road conditions. In [29], sliding mode
control algorithm is also utilized as robust yaw stability and
sideslip control for integrated active front steering and direct yaw
moment. The designed controller was able to track the desired
yaw rate and vehicle side slip with crosswind as an external
disturbance and cornering stiffness as system uncertainty.
Based on the above discussion, integrated active front
steering and direct yaw moment control or differential braking
that use two independent control inputs from two different
actuators i.e. steering and braking could improve the yaw rate and
sideslip tracking control. From the point of view of tracking
control, the transient response performance is very important.
However, the control strategies and algorithms discussed above
does not improve the transient response performances of the yaw
rate and sideslip tracking control in the presence of uncertainties
and disturbances. The designed controllers are only sufficient to
track thedesired responses in the presence of such problems.
Therefore, an appropriate control strategy that could improve the
transientresponse performance for robust tracking control of
vehicle yaw rate and sideslip shouldbe designed and developed
for integrated active steering and braking control.
5.0
HIGH PERFORMANCE ROBUST TRACKING

CONTROL STRATEGY
In previous sections, the control algorithms to ensure a robustyaw

stability controller designwere discussed. In this section, a
proposed robust tracking control strategy with high performance
based on composite non-linear feedback (CNF) control and
sliding mode control (SMC) is discussed in brief.
5.1Composite Nonlinear Feedback Control
The composite nonlinear feedback (CNF) control technique that is
based on state feedback law could improve the transient
performances as designed and implemented in [34] for tracking
control of 2nd order linear system and higher order MIMO linear
system as implemented in [35]. It was extended and generalized
for linear system with actuator nonlinearities in [36], general
multivariable system with input saturation in [37], hard disk drive
servo system, servo positioning system with disturbance in [3841] and is continuously developed and applied in last decade.
109
As in the above studies, the use of CNF control could

improve the performance of transient response, especially for a
second order system based on variable damping ratios. For most
practical situation, it is desired to obtain a fast response with small
overshoot, but in fact, most of control scheme makes a trade-off
between these two transient performance parameters. Hence, the
CNF control technique keep low damping ratio during transient
and varied to high damping ratio as the output response closed to
the set point as illustrated in Figure 4.
Figure 4Transient response improvement using a CNF control technique

[36]
system, fuzzy and neural network system, electrical motor and

drive, power electronics converters, magnetic bearings, advanced
robotics, automotive vehicles and underwater vehicles are
discussed in [42-44].
Sliding mode control design consists of two important steps
i.e. the design of a sliding surface and the design of the control
law so that the system states are enforced to the sliding surface.
The design of the sliding surface is very important as it will
determine the dynamics of the system being controlled. A
conventional linear sliding surface of SMC has a disadvantage in
improving transient response performance of the system due to
constant close loop damping ratio. Thus, a nonlinear sliding
surface that changes a closed loop system damping ratio to
achieve high performance of the transient response, and at the
same time ensure the robustness has been proposed in [45-47].
This concept of varying close loop damping ratio is based on the
composite nonlinear feedback (CNF) control as discussed
previously. Recently, this type of nonlinear sliding surface is
extensively applied for SMC design in current research works
[48-50].
Based on tracking error, the nonlinear sliding surface using a
nonlinear function of CNF for yaw stability control system can be
described as follows
s( z, t ) : c T e(t ) c1
To realize this concept of variable damping ratio, the CNF
control law is designed toinclude the linear and nonlinear
feedback laws as described in the following equations;
u uL uN
(15)
u Fx Gr (r , y) B P( x x e )
where F is feedback matrix, G is a scalar matrix, B is input
matrix.If P>0 is a solution of the Lyapunov equation and (r , y)
is a nonlinear function which is chosenas follows;
(r , y) e ( y r )
( r , y ) e
(r , y )
(16)
y r
1 e 1
(17)
(e
(1( y y0 ) /(r y0 )) 2
e )
(18)
The nonlinear function (r , y) is used to change the

damping ratio of closed loop system from low to final high value
as the output approaches the reference r. The parameters and
are appropriate positive performances i.e.tuned to yield a
desired performances giving fast settling time and small
overshoot.
Although the CNF control technique has an advantage in
improving transient response, its control structure could not cater
to the system with uncertainties and varying disturbances.
Therefore, to make the CNF robust without changing its structure,
it could be combined with sliding mode control for high
performance robust tracking control.
5.2Sliding Mode Control with Nonlinear Sliding Surface
Sliding mode control (SMC) algorithm that had been developed in
the last two decades is recognized as an effective robust controller
to cater to the matched and mismatched uncertainties and
disturbances for linear and nonlinear system. It is also utilized as
an observer for estimation and identification purposes in
engineering system. Various applications using SMC such as in
motion control system, servo system, continuous or discrete
e (t )
I m 1
e 2 (t )
(19)
where
T
c1 : F (r , y)a12
P
(20)
with e1 (t ) and e2 (t ) representingthe yaw rate and sideslip

tracking error respectively. Notice that the nonlinear function
(r , y) can be chosen from equations (16 - 18) and I m is the
identity matrix.
Based on the above discussion, the SMC with nonlinear
sliding surface based on CNF technique could achieve high
performance with the robustness for uncertain systems. It could
improve the transient response performance in the presence of
uncertainties and external disturbances. In addition, it is found
that this control strategy is not yet been examined for vehicle yaw
stability control system and should be further investigated.
Therefore, this control technique has initiated a motivation to
implement with integrated active steering and braking,a robust
yaw rate and sideslip tracking control.
6.0 CONCLUSION
The main elements of vehicle yaw stability control system have
been reviewed. Vehicle models, control objectives, control
problems and control algorithms for integrated active steering and
braking control have been discussed and presented. It is observed
that the designed controllers is not capable to improve the
transient response of the yaw rate and sideslip tracking control,
especially in the presence of uncertainties and disturbances where
theyare only sufficient to track the desired responses. It is also
found thata control strategy using the sliding mode control
algorithm with nonlinear sliding surface based on CNF control
technique is proposed. It is expected that such control strategy
could be implemented for integrated active steering and braking
control in order to achieve high performances for robust tracking
control of vehicles yaw rate and sideslip, improvingthe vehicle
handling and stability.
Acknowledgement
110
The authors are grateful to the Ministry of Education of Malaysia,

UTeM, UTM, and UniKL for the support to conductthis research.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
S.C. Baslamisli, I. E. Kse, G. Anla.2009.Gain-scheduled integrated

active steering and differential control for vehicle handling
improvement.Vehicle System Dynamics.47(1): 99119.
R. Karbalaei, M. Ali, S. Hamed Tabatabaei, R. Kazemi, A. Ghaffari.
2008.Integrated control of AFS and DYC in the vehicle yaw stability
management system using fuzzy logic controlSAE Technical Papers,
SAE World Congress, Michigan, USA,2008-01-1262.
Y. Wu, D. Song, Z. Hou, X. Yuan. 2007.A fuzzy control method to
improve vehicle yaw stability based on integrated yaw moment control
and active front steering.Proceedings of IEEE International Conference
on Mechatronics and Automation (ICMA 2007). Harbin, China.1508
1512.
S. C.Balamili, I. E Kse, G. Anla.2006.Design of active steering and
intelligent braking systems for road vehicle handling improvement: A
robust control approach.Proceedings of IEEE International Conference
on Control Applications (CCA 2006). Munich, Germany.909914.
P. Falcone, H. Eric Tseng, F. Borrelli, J. Asgari, D.Hrovat. 2008.MPCbased yaw and lateral stabilisation via active front steering and
braking.Vehicle System Dynamics.46(1):611628.
J. He, D. A. Crolla, M.C Levesley, W. J Manning. 2006.Coordination of
active steering, driveline, and braking for integrated vehicle dynamics
control," proceedings of the Institution of Mechanical Engineers. Part D:
Journal of Automobile Engineering, 220(10):14011421.
S. B. Lu,Y. N Li, S. B. Choi, L. Zheng, M. S. Seong.2011.Integrated
control on MR vehicle suspension system associated with braking and
steering control.Vehicle System Dynamics.49(12):361380.
J. Wu, Q. Wang, X.Wei, H. Tang. 2010.Studies on improving vehicle
handling and lane keeping performance of closed-loop driver-vehicle
system with integrated chassis control.Mathematics and Computers in
Simulation.80(12):22972308.
N. Ding, S. Taheri. 2010.An adaptive integrated algorithm for active
front steering and direct yaw moment control based on direct Lyapunov
method.Vehicle System Dynamics.48(10):11931213.
X. Yang, Z.Wang, W. Peng. 2009.Coordinated control of AFS and DYC
for vehicle handling and stability based on optimal guaranteed cost
theory.Vehicle System Dynamics.47(1):5779.
B.Lee, A. Khajepour, K. Behdinan.2006.Vehicle stability through
integrated active steering and differential braking.SAETechnical Papers,
SAE World Congress,Michigan, USA.2006-01-1022.
M. Selby, W. J.Manning, M. D. Brown, D. A Crolla. 2001.A
coordination approach for DYC and active front steering,SAE Technical
Papers, SAE World Congress, Michigan, USA,2001-01-1275.
S. Moon, W. K.Cho, K. S Yi.2010.Intelligent vehicle safety control
strategy in various driving situations.Vehicle System Dynamics.48:537
554.
R. Marino, S.Scalzi, 2009.Integrated control of active steering and
electronic differentials in four wheel drive vehicles.SAE International
Journal of Passenger Cars - Electronic and Electrical Systems.2(1):141
149.
D. Li, S. Du, F. Yu. 2008, Integrated vehicle chassis control based on
direct yaw moment, active steering and active stabiliser.Vehicle System
Dynamics.46(1):341351.
W. Cho, J. Yoon,J. Kim,J.Hur, K. Yi. 2008.An investigation into unified
chassis control scheme for optimised vehicle stability and
manoeuvrability.Vehicle System Dynamic.46(1):87105.
D. Li, F. Yu.2007.A novel integrated vehicle chassis controller
coordinating direct yaw moment control and active steering.SAE
Technical Papers, proceeding of 14th Asia Pacific Automotive
Engineering Conference, California, USA.2007-01-3642
S. H. Tamaddoni, S. Taheri, M. Ahmadian. 2011.Optimal preview game
theory approach to vehicle stability controller design.Vehicle System
Dynamics.49(12):19671979.
R. Rajamani, Vehicle Dynamics and Control: 2nd Ed, 2011, USA,
Springer, Chapter 2.
Z. He, X. Ji. 2012.Nonlinear robust control of integrated vehicle
dynamics.Vehicle System Dynamics.50(2):247280.
C. Rengaraj, D. Crolla. 2011.Integrated chassis control to improve
vehicle handling dynamics performance.SAE Technical Papers, SAE
2011 World Congress and Exhibition, Detroit, USA.2011-01-0958.
S. C. Baslamisli, I. E Kose, G. Anla. 2011.Handling stability
improvement through robust active front steering and active differential
control.Vehicle System Dynamics.49(5):657683.
[23] J. Wang, R. G. Longoria. 2009.Coordinated and reconfigurable vehicle

dynamics
control.IEEE
Transactions
on
Control
Systems
Technolog.17(3): 723732.
[24] A. Goodarzi, A.Sabooteh, E. Esmailzadeh. 2008.Automatic path control
based
on
integrated
steering
and
external
yaw-moment
control.Proceedings of the Institution of Mechanical Engineers, Part K:
Journal of Multi-body Dynamics.222(2):189200.
[25] H. B. Pacejka, Tyre and Vehicle Dynamics: 2nd Ed, 2006, UK,
Butterworth-Heinemann, Chapter 1 & 4.
[26] C. Poussot-Vassal, O. Sename, L.Dugard, S. M. Savaresi. 2011.Vehicle
dynamic stability improvements through gain-scheduled steering and
braking control.Vehicle System Dynamics.49(10): 15971621.
[27] J. Guo,L. Chu, H. Liu, M. Shang, Y. Fang. 2010.Integrated control of
Active Front Steering and Electronic Stability Program,IEEE
International Conference on Advanced Computer Control(ICACC
2010).449453.
[28] C.Ahn, B.Kim, M. Lee.2012.Modeling and control of an anti-lock brake
and steering system for cooperative control on split-mu
surfaces.International Journal of Automotive Technology.13(4): 571
581.
[29] A. Hu, F. He. 2011.Variable structure control for active front steering and
direct yaw moment.Proceedings of 2nd International Conference on
Artificial Intelligence, Management Science and Electronic Commerce
(AIMSEC 2011), Zhengzhou, China.35873590.
[30] A. Hu, B. Lv.2010.Study on mixed robust control for integrated active
front steering and direct yaw moment.Proceedings of IEEE International
Conference on Mechatronics and Automation (ICMA 2010), Xi'an,
China.2933.
[31] W. J. Manning, D. A. Crolla. 2007.A review of yaw rate and sideslip
controllers for passenger vehicles.Transactions of the Institute of
Measurement and Control.29(2): 117135.
[32] F. Yu, D. F Li,D. A. Crolla.2008.Integrated vehicle dynamics controlstate-of-the art review.Proceedings of IEEE Vehicle Power and
Propulsion Conference (VPPC 2008), Harbin, China.
[33] J. Tjonns, T. A. Johansen.2010.Stabilization of automotive vehicles
using active steering and adaptive brake control allocation.IEEE
Transactions on Control Systems Technology.18(3): 545558.
[34] Z.Lin,M.Pachter, S. Ban.1998.Toward improvement of tracking
performance - Nonlinear feedback for linear systems.International
Journal of Control.70(1): 111.
[35] M. C. Turner, I.Postlethwaite, and D. J. Walker. 2000.Non-linear
tracking control for multivariable constrained input linear
systems.International Journal of Control.73(12): 11601172.
[36] B. M.Chen,T. H.Lee, K. Peng, V. Venkataramanan. 2003.Composite
nonlinear feedback control for linear systems with input saturation:
Theory and an application.IEEE Transactions on Automatic
Control.48(3): 427439.
[37] Y. He, B. M.Chen, C. Wu. 2005.Composite nonlinear control with state
and measurement feedback for general multivariable systems with input
saturation.Systems and Control Letters.54(5): 455469.
[38] W. Lan, C. K.Thum, B. M. Chen. 2010.A hard-disk-drive servo system
design using composite nonlinear-feedback control with optimal
nonlinear gain tuning methods.IEEE Transactions on Industrial
Electronics.57(5): 17351745.
[39] G. Cheng, K. Peng. 2007.Robust composite nonlinear feedback control
with application to a servo positioning system.IEEE Transactions on
Industrial Electronics.54(2): 11321140.
[40] G.Cheng, W. Jin. 2006.Parameterized design of nonlinear feedback
controllers for servo positioning systems.Journal of Systems Engineering
and Electronics.17(3): 593599.
[41] K. Peng, B. M.Chen, G. Cheng, T. H. Lee.2005.Modeling and
compensation of nonlinearities and friction in a micro hard disk drive
servo system with nonlinear feedback control.IEEE Transactions on
Control Systems Technology.13(5): 708721.
[42] Wilfrid Perruquetti, Jean Pierre Barbot.2002. Sliding Mode Control in
Engineering. New York, USA, Marcel Dekker Inc.
[43] Hocine Imine, Leonid Fridman, Hassan Shraim, Mohamed Djemai.2011.
Sliding Mode Based Analysis and Identification of Vehicle
Dynamics.Berlin, Germany, Springer.
[44] Vadim Utkin, Jrgen Guldner, Jingxin Shi. 2008.Sliding Mode Control in
Electro Mechanical Systems, 2nd Ed.New York, USA, CRC Press.
[45] B.Bandyopadhyay, D.Fulwani, I.Postlethwaite, M. C.Turner.2010.A
nonlinear sliding surface to improve performance of a discrete-time
input-delay system.International Journal of Control, 83(9): 18951906.
[46] B.Bandyopadhyay, D. Fulwani. 2009.A robust tracking controller for
uncertain MIMO plant using non-linear sliding surface.Proceedings
ofIEEE International Conference on Industrial Technology, ICIT,
Churchill, USA.
111
[47] Bijnan Bandyopadhyay, Kyung-Soo Kim, Fulwani Deepak.2009. Sliding

Mode Control Using Novel Sliding Surfaces.Berlin, Germany, Springer.
[48] S. Mondal, C. Mahanta.2012.A fast converging robust controller using
adaptive second order sliding mode.ISA Transactions.51(6): 713721.
[49] S. Mobayen, V. J. Majd, M. Sojoodi.2012.An LMI-based composite
nonlinear feedback terminal sliding-mode controller design for disturbed
MIMO systems.Mathematics and Computers in Simulation, 85: 1-10.
[50] S.Mobayen, V. J. Majd, M. Sojoodi. 2012.An LMI-based finite-time
tracker design using nonlinear sliding surfaces.Proceedings of
20thIranian Conference on Electrical Engineering, (ICEE 2012), Tehran,
Iran,810815.
2015 International Conference on Unmanned Aircraft Systems (ICUAS)

Denver Marriott Tech Center
Denver, Colorado, USA, June 9-12, 2015
A Review on the Platform Design, Dynamic Modeling and Control of

Hybrid UAVs
Adnan S. Saeed, Ahmad Bani Younes, Shafiqul Islam, Jorge Dias, Lakmal Seneviratne, Guowei Cai
Khalifa University of Science, Technology and Research
Abu Dhabi, UAE
{adnan.saeed, ahmad.younes, shafiqul.islam, jorge.dias, lakmal.seneviratne, guowei.cai}@kustar.ac.ae
Abstract This article presents a review on the platform
design, dynamic modeling and control of hybrid Unmanned
Aerial Vehicles (UAVs). For now, miniature UAVs which have
experienced a tremendous development are dominated by two
main types, i.e., fixed-wing UAV and Vertical Take-Off and
Landing (VTOL) UAV, each of which, however, has its own
inherent limitations on such as flexibility, payload, axnd endurance. Enhanced popularity and interest are recently gained
by a newer type of UAVs, named hybrid UAV that integrates the
beneficial features of both conventional ones. In this paper, a
technical overview of the recent advances of the hybrid UAV is
presented. More specifically, the hybrid UAVs platform design
together with the associated technical details and features
are introduced first. Next, the work on hybrid UAVs flight
dynamics modeling is then categorized and explained. As for
the flight control system design for the hybrid UAV, several flight
control strategies implemented are discussed and compared in
terms of theory, linearity and implementation.
Fig. 1.
the miniature hybrid UAVs to become more mature, in terms

of design philosophy, dynamics modeling, control, guidance,
navigation, robustness, and etc.
The main contribution of this paper is to present an
overview on the recent advances of the hybrid UAVs, mainly
concentrating on platform design, flight dynamics modeling
and control methods. It is the authors desire that this work
provides a fairly complete picture of the mini hybrid UAVs
and serves as a baseline for further developments of this field.
I. I NTRODUCTION
During the last two to three decades, Unmanned Aerial
Vehicles (UAVs) have experienced a tremendous development. For now, miniature UAV platforms are dominated by
two main types, i.e., fixed-wing conventional aircraft and
Vertical Take-Off and Landing (VTOL) aircraft, and each
type has its own inherent limitations on such as flexibility,
payload, endurance, and etc. A new and promising trend is
to develop fixed-wing VTOL UAV or the so-called Hybrid
UAV, which can inherit the advantages of both and thus have
the ability of vertical take-off and landing as well as high
cruising speed and enhanced endurance. This enables the
possibility of performing wider range of missions or same
missions with better performance.
Indeed, integrating the advantages of fixed-wing and
VTOL aircraft has long been a concern for many aerospace
and aviation industries. Over the years, there have been
several attempts to build manned hybrid aircraft such as
Bell Boeing V-22 Osprey, Vertol VZ-2, Sikorsky X-wing,
Convair XFY Pogo and Harrier GR7 as shown in Fig. 1(a)
to 1(e), respectively [1][2][3][4][5]. Some of the attempts did
succeed and the aircraft are still operating up to the moment
such as V22-Osprey and Harrier GR7. Nevertheless, within
the last four years, the concept invaded the UAV field as a
number of research groups documented their pioneer work
in literature and a couple of companies even commercialized
the idea. It is believed that the hybrid UAVs will be having
a bright future and will promptly dominate the miniature
UAV market. Still in its infancy, there is a huge space for
978-1-4799-6009-5/15/$31.00 2015 IEEE
Examples of Manned Hybrid Aircraft.
II. P LATFORM D ESIGN

Hybrid UAVs can be generally categorized into two main
types: Convertiplanes and Tail-Sitters. Each of them can be
further categorized into a few sub-types, depending on the
transition mechanism and airframe configuration.
A. Convertiplane
A convertiplane is a type of hybrid aerial vehicle that takes
off, cruises, hovers and lands with the aircrafts reference line
remaining horizontal (i.e., the main body configuration does
not change during flight). A variety of transition mechanisms
are applied to achieve the conversion from vertical flight to
horizontal flight and vice versa. Based on that, convertiplanes
can be further categorized into four sub-types, including 1)
Tilt-Rotors, 2) Tilt-Wings, 3) Rotor-Wings, and 4) DualSystems.
1) Tilt-Rotor: The primary feature of the Tilt-Rotor aerial
vehicle is that the multiple rotors used are mounted on
rotating shafts or nacelles. During transition, the rotors
tilt gradually towards flight direction providing the aircraft
forward speed until level flight mode is achieved. Some TiltRotors might have fixed rotors always directed upwards and
806
Fig. 2.
utilize any coaxial rotor because the differential thrust can

be canceled using rudder deflections [6] [7]. It is also worthy
to note that, although a tilt duct VTOL UAV concept is
presented in [14], no actual platform was built by the authors.
2) Tilt-Wing: A Tilt-Wing has a similar concept to TiltRotor except that the assembly of the wing tilts instead of
the rotors only. During takeoff, landing and hovering, the
wings will be directed upwards which makes the aircraft
more vulnerable to cross winds. Consequently, the TiltWing requires complicated control mechanisms and higher
use of available power to maintain stability during vertical
flight. Furthermore, landing on moving deck environments is
relatively more difficult compared to Tilt-Rotors. However,
since rotors are fixed to wings, this allows various design
options for the wing geometry and therefore enhance the
aerodynamic performance of the aircraft. Vertol VZ-2 (Fig.
1b) was a manned Tilt-Wing aircraft built in 1957 by Boeing
Vertol, completed several successful flights and was retired
in 1965 [2]. Like Tilt-Rotors, Tilt-Wings are also researched
actively and there are many examples applying the idea such
as AVIGLE (Fig. 3(a)) [15][16][17], Sabanci University UAV
(SUAVI) (Fig. 3(b)) [18] [19], AT-10 Responder (Fig. 3(c))
[6], Quad Tilt Wing (QTW) VTOL UAV [20], and HARVee
[21].
Due to the requirement of a wing tilting mechanism,
conventional configuration design is essential for Tilt-Wings
and was utilized in most reviewed works with single main
wing favored in AT-10 and AVIGLE and a tandem wing
design favored in SUAVI and QTW VTOL UAV. It was
found that most of the works implementing the quad TiltWing hybrid UAVs have all the rotors tilt to make the
transition to forward flight. Such configuration might not
be optimal because two tilting mechanisms will have to
equipped, one for the front wing and another for the rear
wing, which increases weight as well as the complexity of
the design. Moreover, the rear rotors do not contribute much
in increasing the forward thrust in level flight, instead the
increased drag might as well outweighs the extra forward
thrust provided. Therefore, an improvement would be to let
the front wing tilt and keep the rear rotors directed upwards
(only rotors, not wings) to operate only during vertical flight.
On the other hand, the controllability and stability of the
single wing Tilt-Wings during vertical flight is somehow
questionable because any perturbation that causes a sudden
change in the angle of attack will not be countered by any
control surface or differential thrust actuation.
3) Rotor-Wing: A Rotor-Wing (or Stop-Rotor) is another
type of Convertiplane aircrafts where rotary wings spin to
provide lift during vertical flight and stop to act like a fixed
wing during horizontal flight. Sikorsky X-wing (Fig. 1c)
is a manned aircraft implementing the Rotor-Wing concept
but the program was cancelled in 1988 because funds ran
out after only three flights [5] [22]. Boeing X-50 Dragon
Fly (Fig. 3(d)) developed by DARPA is another Rotor-Wing
UAV but was also withdrawn because of some aerodynamic
problems [23]. The way the Rotor-Wings work makes the
possible design configurations minimal which is why most
Examples of Tilt-Rotor UAVs.
operate only during vertical flight to provide extra lift for

takeoff, landing and hovering. An example of manned TiltRotor is Bell Boeing V-22 Osprey shown in Fig. 1(a) which
has two tilting jet engines to perform the transition [1]. As in
V-22 Osprey, Tilt-Rotors usually have their engines mounted
at the wing tips forcing shorter wing span and thicker
airfoil. This results in lower aspect ratio and increased drag
respectively causing poor aerodynamic performance. On the
other hand, the shafts and nacelles are only required to rotate
rotors instead of wings or other heavy structures which saves
power and weight. Moreover, due to their controllability and
stability in vertical flight when compared to other hybrid
UAVs, Tilt-Rotors are actively researched in academia and
there exist several vehicles implementing the idea such as
IAI Panther [6][7], TURAC [8][9] [10], Orange Hawk [11],
FireFLY6 [12] and AgustaWestland Project Zero [13] shown
in Fig. 2(a) to 2(e), respectively.
The most dominant design for Tilt-Rotors is the flying
wing configuration as this was utilized in most reviewed
works. However, a conventional airplane configuration was
also implemented in IAI Panther and Mini Panther. The
conventional design is easier to build, analyze and manufacture whereas the flying wing design has a larger wing
area to generate lift and therefore better payload capacity.
Regarding the propulsion system characteristics, a tricopter
configuration is the most common one implied in both
configurations which is mainly due to the cost benefit over
other multi-copters and the simplicity of the forward flight
control system. Also, the tricopter configuration makes the
center of gravity of the airplane to be near the center of lift of
the tricopter configuration which further enhances stability
and controllability [11]. All the works reviewed utilized
tricopter configuration except for AgustaWestland Project
Zero [13] which has only two tilting rotors that are located in
large holes in an otherwise conventional flying wing aircraft.
However, few technical details and characteristics for this
specific aircraft were found in [13]. In the tricopter design,
the differential thrust resulting from the rear motor causing
unstable motion can be resolved using a rear coaxial rotor
as in TURAC [8] [9] [10] and Orange Hawk [11] or by
using all three coaxial rotors as in FireFLY6 [12]. However,
IAI Panther with the conventional configuration does not
807
Fig. 3.
of the projects was successful [4][30]. Tail-sitters can be

classified into three types as follows.
1) Ducted-Fan UAVs: Ducted-Fan VTOL UAV is a type
of tail-sitter UAVs where a large duct fan usually coaxial
forms the main body of the aircraft and several control
surfaces are installed for stability, control and transition. The
complexity of the control and stability strategies, payload
capacity, range, endurance and cruising speed are some
drawbacks for this type of design. Several works presented
this concept in literature and it was found that this type of
UAVs was mainly used for military purposes. RMIT University ducted-fan aimed to aid law enforcement activities [31]
and Vertical Bat developed by Brigham Young University
partnering with MLB Company [32] are some examples but
no prototype was tested in the full flight envelope. Bertin
Hovereye and Selex Galileo Asio ((Fig. 4(a)) are other
examples that have been marketed since 2005 and 2008 [6].
2) CSTTs and DTTTs: Apart from Ducted-fans, other
tail-sitters can be classified into those performing the transition using large control surfaces and those performing
that using differential thrust from their multiple rotors. The
former can be referred to as Control Surface Transitioning
Tail-sitters (CSTT) and the latter as Differential Thrust
Transitioning Tail-sitters (DTTT). Both require complicated
control mechanisms and higher use of available power to
maintain stability during vertical flight because they are
more susceptible to cross winds which also makes landing
on moving decks difficult. CSTTs usually require only a
single or sometimes twin rotors. In contrast, DTTTs require
multiple rotors to enable providing sufficient differential
thrust to make transitions which results in reduced efficiency
in horizontal flight. However, in terms of stability, DTTTs
are far more stable in takeoff, hovering, and landing and
require simpler control strategies than CSTTs. ITU Tailsitter with a folding propeller system located on its nose
that operates during vertical flight and an electric ducted fan
located in the tail that operates during horizontal flight with
the propeller blades folded to reduce drag [33] and T-Wing
((Fig. 4(b)) with canard wing and tandem rotors developed
by Stone and his colleagues and used for defense and
civilian applications [30] [34] are some examples of CSTTs.
As mentioned earlier, these tailsitters are based on a fixed
wing conventional airplane design configuration and their
attitude is controlled by the control surfaces which results
in high stability during horizontal flight but low stability and
controllability during vertical flight. VertiKUL ((Fig. 4(c))
[35], ATMOS ((Fig. 4(d)) [36] and Quadshot [37] [38] are
examples of DTTTs. All three have four rotors to provide
differential thrust to make the transition to horizontal flight
and back. However, ATMOS and Quadshot have control
surfaces or tilting rotors for control during horizontal flight
unlike VertiKUL which depends solely on differential thrust.
3) Reconfigurable wings: Reconfigurable wings are another type of Tail-sitters where the wings extend during
horizontal flight and retract during vertical flight. This concept is based on the idea that more lift at lower speed
is desired in cruising and therefore the wings extend to
Examples of Convertiplane UAVs.
of the Rotor-Wing works show similar design configuration

which is close to a helicopter. However, a different design
configuration in which the whole body tilts during transition
was detailed in [24] and [25] but no full envelope powered
flight test was conducted. Therefore, the implementation of
such design might be suspicious due to the lack of historical
successes.
4) Dual-Systems: Another type of Convertiplanes which
could be referred to as Dual-System implements multiple
rotors always directed upwards for vertical flight and another
separate tractor or pusher for level flight. Many possible
design configurations are possible for this specific type of
UAV but due to the huge weight of the aircraft, quad
rotors are the most common. The concept of Dual-Systems
is very simple to apply in terms of design, controllability,
stability and modeling because the two flight modes could
be analyzed separately. However, during horizontal flight,
the multiple lifting rotors used for vertical flight are not in
operation and add extra weight to the aircraft which results
in requiring more power from the tractor or pusher. From
the survey conducted, it was found that the idea is employed
in Arcturus JUMP (Fig. 3(e)) developed by Arcturus UAV
[26] [27] and in Airbus Quadcruiser, shown in Fig. 3(f) [28]
[29]. Both designs are quadcopters with a tractor in Arcturus
JUMP and a pusher in Quadcruiser. A proof-of-concept 1.50
m wingspan Airbus Quadcruiser successfully performed its
maiden flight to test the aircrafts stability, controllability and
handling while hovering and low speed cruising [29].
B. Tail-sitter
A Tail-sitter is an aircraft that takes off and lands vertically
on its tail and the whole aircraft tilts forward using differential thrust or control surfaces to achieve horizontal flight.
This concept could also be denoted as Tilt-Plane since the
whole plane tilts to achieve level flight. Due to its ability to
make the transition without the need of extra actuators, this
concept is mechanically simple and saves a huge amount of
weight when compared to Convertiplanes. Moreover, since
tail-sitters land on their tails, they require relatively stronger
tails to be able to withstand landing impacts. In the last 50
years, Tail-sitters have been analyzed extensively and there
were several trials to build manned ones such as Convair
XFY Pogo (Fig. 1(e)) and Lockheed XF-V1. However, due
to the difficulty of control during transition and landing, none
808
Fig. 5. Block diagram of the flight dynamics model of the hybrid aerial
vehicles.
Fig. 4.
TABLE I
R EPRESENTATIVE MODELING
Aircraft Type
key components are contained. Detailed explanations of these

components are given as follows.
First, as the hybrid UAVs are only capable of operating
in a relatively confined area, two main Cartesian coordinate
systems, i.e., local coordinate (mainly north-east-down) and
body-frame coordinate, are generally sufficient for describing
the kinematic and dynamic motions of hybrid aerial vehicles.
For the level flight mode, the stability coordinate is additionally taking into account the wind effect. The definitions of
all these coordinate systems can be easily found in variety
of textbooks (see, e.g., [66] and [67]).
With the coordinate systems defined, the kinematics, i.e.,
orientation and translation relations w.r.t. the operational
environment, of hybrid aerial vehicles can be determined.
The former can be expressed via two formulations, namely,
Euler angles and Quaternion, depending on the specific type
of the hybrid vehicles as well as the specified missions.
Detailed explanations of both expressions can be found in
textbooks such as [66]. Generally, Euler angle representation
dominates the convertiplanes and single mode (e.g., hover
or cruise with constant speed) of partial tail-sitters because
the fuselage does not change with large amplitude. On the
other hand, the tail-sitters operating over full flight envelope commonly adopt quaternion representations because the
transition between hover and level flight modes leads to approximately 90-degree pitch angle change which has a high
chance to raise the singularity of Euler-angle representation.
Representative implementation of quaternion expression on
hybrid aerial vehicles will be addressed later in Section III-B.
The second component to be addressed is the rigid body
dynamics which concerns the translational and rotational
equations of the hybrid vehicles. Generally, Newton-Euler
and Euler-Lagrange formulations, which involve the combined forces F and moments M, are employed to achieve this
aim. It should be noted that, corresponding to the different
orientation expressions (i.e., Euler angle or Quaternion), the
expressions of F and M differ accordingly. On the other
hand, Euler-Lagrange approach does not require a particular
identification of the coordinate system and makes use of the
conservation of energy to derive the equations of motion
instead. Thus, the translational and rotational equations of
motions could be derived from Euler-Lagrange equation
which involves the generalized coordinates of the system q
and the combined forces and moments F w.r.t. the C.G of
a hybrid UAV. According to our survey, only few research
works (i.e., [58], [59], [60]) adopt this formulation without
Examples of Tail-sitter UAVs.
RESEARCH WORK .
Representative Modeling Work
Tilt-Rotor
[10], [41], [42], [43], [44], [45], [46], [47], [48], [49]
Tilt-Wing
[15], [19], [20], [50], [51]
CSTT
[32], [52], [53], [54], [55], [56], [57], [58], [59], [60],
[61], [62], [63], [64], [65]
DTTT
[35]
provide larger wing span. Moreover, the wings retract during

vertical flight to minimize the effect of wind disturbance
and ease maneuverability. The idea is applied in U-Lion
((Fig. 4(e and f)), an aircraft developed by the National
University of Singapore in 2014 in which a four bar linkage
was designed to make the reconfiguration which in turn
minimizes platform weight and keeps the structure simple
[39]. [40] also presents the dynamics model of HADA which
performs the configuration by unfolding the wings beneath
the fuselage. However, no flight experiment was held yet.
III. F LIGHT DYNAMICS M ODELING
A reliable model that can accurately capture the flight
dynamics over the flight envelope of interest is critically
important for developing the autonomous flight control
system. For the hybrid UAVs, the flight envelope can be
generally divided into three modes, namely, vertical flight
mode, transition mode, and level flight mode. As a result,
developing a reliable flight dynamics model becomes more
challenging compared with any conventional aircraft. In the
recent literature, a number of flight modeling works for
the hybrid UAVs have been documented. The representative
work are listed in Table I for convenient reference. In what
follows of this section, we provide a brief overview of the
flight dynamics modeling work via two aspects: 1) model
structure of the hybrid UAV dynamics, and 2) representative
modeling work.
A. Model Structure
Our survey indicates that the flight dynamics of the hybrid
UAVs can be uniformly depicted in Fig. 5, in which six
809
highlighting the particular reason for their selection.

Aerodynamic forces (lift and/or drag) and the associated
moments are constantly generated by the various control
surfaces and the fuselage of hybrid aerial vehicles in operation. Compared with the conventional fixed-wing or rotary aircrafts, hybrid vehicles dynamic modeling is more
complex as the aerodynamic features of the aforementioned
components in different flight conditions or envelopes should
be taken into account comprehensively. A prevalent method
dominantly adopted by the documented works is to express
the aerodynamic forces and moments using dimensionless
coefficients which are determined via practical flight experimental data as in [62], wind tunnel experimental data as in
[18], [19], [49] or CFD results as in [10], [48], [52].
Regarding the dynamics of the propulsion system, two
coupled sub components (i.e., the propeller aerodynamics
and the motor dynamics) are involved. For the former, the
majority of the modeling works on hybrid aerial vehicles
adopt highly simplified dynamics model which involves very
fundamental aerodynamic analysis. For instance, in [62],
[63] only quasi-steady equations are utilized to model the
aerodynamic forces and moments. For the latter, according
to our survey, no research work has particularly paid attention
to the motor dynamics. Thus, the response of propulsion
systems to the actuator input is assumed instantaneous.
Finally, the titling mechanism uniquely belongs to convertiplane aircraft. Thus, in Fig. 5 it is put in a dashed
block. Similar to the propeller aerodynamics part, the current
hybrid aerial vehicle modeling works adopt highly simplified
models to account for the titling motion of the propulsion
systems. For instance, a common method has been documented in [41], [43], [68] in which two instantaneous shaft
tilting angles, L and R , are defined for the rotation of
the left and right front propulsion systems and the tilting
motion is reflected by a rotation matrix based on tilting
angles defined.
systems (such as Nacelle axis system, hub-axis system, and

blade axis system) and including the flapping motion of
the propellers. Furthermore, the aerodynamics of the control
surfaces and fuselage are carefully determined via variety
of wind-tunnel experiments. Model validation in both timeand frequency-domains is presented and the results indicate
the relative high fidelity of the proposed model. In another
work documented in [47], the essential role of the wind
tunnel usage in determining various aerodynamic coefficients
is clearly demonstrated via both large amount of data and
model validation results. Instead of using the experimental
results collected in the wind-tunnel, the authors of [10], [48]
have explored the possibility of using CFD to determine
the aerodynamic coefficients for a 0.15-scale MV-22 birotor convertiplane and a custom-built tri-rotor convertiplane
TURAC respectively. Partial validation results have also been
presented in [48] to prove the efficiency of the CFD-based
estimation.
Compared with tilt-rotor hybrid aircraft, less interest in
modeling tilt-wing hybrid aircraft has been observed. Furthermore, [19], [50], [51] are based on an identical custombuilt miniature quad tilt-wing hybrid UAV and only one
modeling work on twin tilt-wing hybrid aircraft [15] has
been found. All the proposed models adopt Euler-angle
expression, Newton-Euler formulation, and highly simplified
motor dynamics and tilting mechanisms as mentioned in
Section III.A. One distinguished feature shared by all the
documented work on tilt-wing hybrid aircraft listed in Table
I is that wind tunnels are uniformly used to determine the
aerodynamic coefficients involved in the developed models.
A number of research works on CSTTs, which is either
single-rotor- or bi-rotor-based, have been carried out and
documented in literature. Part of them only focuses on
vertical flight mode and attitude stabilization. For instance,
in [58], [59], [60], [61], two types of bi-rotor CSTTs have
been developed and Euler-Lagrange formulation is adopted
in modeling their dynamics. As attitude stabilization in hover
model is the focus, Euler angle instead of quaternion is used
for more straightforward attitude representation. The rest of
the modeling works for CSTTs address the dynamics model
covering the full envelope, that is, hover, transition, and level
flight. More specifically, [32], [52], [54], [56], [56], [63], [65]
concentrate on the modeling of a single-rotor hybrid aerial
vehicles. All these works adopt 1) quaternion formulation
for avoiding the singularity in pitch angle expression and
2) simple expression for propulsion systems. In order to
enhance the accuracy of the proposed model, additional effort
has been made in some documented works, mainly on motor
dynamics and aerodynamic coefficients determination. For
instance, in [52], ducted-fan design code is employed to
account for the unique duct fan feature of the custom-made
CSTT developed at the KAIST and Navier-Stokes solver
integrated in FLUENT toolkit is used to determine aerodynamic control coefficients. In another two documentations
[54], [65] based on a miniature single-rotor hybrid UAV
developed at BYU, aerodynamic coefficients are determined
by maximally matching the flight test data collected in
B. Representative Modeling Work

Table I provides a complete list of the documentations
related to the dynamics modeling of hybrid aerial vehicles
following the categorization method introduced in Section
II. According to review conducted, no systematic modeling
work on 1) rotor/wing, 2) dual system, and 3) reconfigurable
hybrid aircrafts have been documented in the literature, thus
in Table I only the remaining four sub-categories are listed.
The focus of this section is to analyze some unique features
of the modeling work given in Table I.
Starting with the tilt-rotor hybrid aerial vehicles, most of
the works (e.g., [41], [42], [43], [68] for bi-rotor convertiplane, [44], [49] for tri-rotor convertiplane, and [46] for
quad-rotor convertiplane) employ highly simplified motor
dynamics and titling mechanisms to minimize the complexity
of the overall model. An exception that can be treated as
a benchmark is the modeling work documented in [49], in
which a fairly complete flight dynamics model for a bi-rotor
convertiplane has been proposed. The propulsion system
is modelled in depth by introducing additional coordinate
810
experiments. Furthermore, [54] also addresses a technique of

modeling the angular dynamics as a combination of one bias
acceleration term and one actuator-based input term which
aims at reducing the computational load of physical parameter estimation. However, except the work documented in
[56] which provides identification results for the longitudinal
motions, none of the aforementioned works have included
results on model fidelity validation. On the other hand,
research works documented in [55], [57], [62], [64] focus on
the modeling of bi-rotor hybrid aerial vehicles. Quaternion
expression is dominantly adopted and certain unique features
such as variable pitch propeller [69] and motor dynamics
[57], [64] are additionally considered aiming at covering
the key dynamic features. Validation results and analysis
are again rarely addressed with the exception of [62], in
which a comparison between the model responses and actual
experimental data is conducted and non-ignorable deviations
have been observed for all channels which indicates that the
model accuracy can be further enhanced.
For DTTT, very rare work on dynamical modeling has
been documented in the literature, as DTTT-based UAV is
still a relatively new topic to the academia and very less
systematic research has yet been conducted. One representative work on DTTT dynamics modeling is presented in
[35], in which a quaternion-based Newton-Euler formulation
model is proposed for the custom-made quadcopter tailsitter named VertiKUL. However, no detailed description
of the aerodynamic coefficients determination and propeller
dynamics is contained, which makes the validation of model
fidelity difficult.
attitude controller directly linked to the control surfaces of

the UAV. It gets the reference commands from a mid-level
velocity controller which gets the reference commands from
the high-level position controller [70]. This control theory
was implemented in Ducted-Fan Tail-sitters as detailed in
[32], [52], [72], [73], Tilt-Rotors and Tilt-Wings as detailed
in [18], [19], [14], [74], [75] and other tail-sitters, mainly
for hovering, as detailed in [35], [56], [76].
For the classical control theory, the reference commands
are saturated through the outer loops prior to reaching the
innermost loop which mean that this approach is it results
in good handling of the flight variables and actuator inputs.
Moreover, every single control loop design is simple as it
involves few variables and sometimes even single variable.
However, the difficulty arises when decomposing and separating the variables for the controllers and also when linking
the controllers together especially in terms of determining
whether the inner loops are faster than the outer ones. The
design algorithms of this classical approach depend on one
loop at a time and are very effective for Single-Input SingleOutput (SISO) systems. However, complex systems such as
hybrid UAVs are Multi-Input Multi-Output (MIMO) systems.
Therefore, the success of the classical control system when
applied to hybrid UAVs is not guaranteed because it comprises multiple actuators such as elevators, rudders, ailerons,
motors and tilting rotors and multiple sensors for attitude,
velocity, and position [66] [69] [70] [77].
The second approach, known as modern control, is to
design a control system that handles the full dynamics of
the UAV. The stability and control specifications can be
expressed in terms of a system of first-order differential
equations which results in matrix equations that can be
solved using commercially available computer software to
compute the control gains simultaneously [66] [69] [77]. This
means that all the feedback loops are closed at the same
time. Therefore, a better performance for MIMO systems is
achieved compared to the classical approach in which the
control gains are selected individually. This quick and direct
modern control approach can be utilized for time varying
and time invariant systems, whereas, the classical approach
is mainly for time invariant systems [78]. Also, there are
many optimal control techniques that could be applied in
the modern control theory to improve the controllability and
stability of the UAV [72]. However, not all states correspond
directly to a single actuator as was the case for the first
approach. Therefore, it is difficult to handle actuator saturation [70]. However, modern control theory is also popular in
hybrid UAVs as it was employed in Tilt-rotors as described
in [41], [42], [68], [79], Tail-sitters as detailed in [54], [57],
[76], [80] and Ducted-Fans as detailed in [52], [73], [14].
IV. H YBRID UAV F LIGHT C ONTROL T ECHNIQUES

The core of the control system depends on the derived
dynamics model. As seen in Section III of the paper, the
equations of motion are highly complicated and nonlinear.
Particularly speaking, the dynamics of the hybrid UAVs can
be inherently unstable because it inherits the operation of a
fixed-wing and VTOL UAVs. Even if horizontal and vertical
modes were analyzed separately, the transition phase remains
a critical part of the control system due to the multiple
nonlinearities in the model. That is why feedback control is
essential which ensures more accurate and quicker response
to meet the desired reference command.
A. Flight Control System Theory
Flight control system theory deals with the synthesis and
analysis of the logic behind which the flight control system
is designed. There are two strategies for that, namely, the
classical control theory and the modern control theory. The
former, also known as successive loop closure, considers
the decomposition of the states derived from the model to
form successive control loops such that the output of the
innermost loop (low-level) is linked to the actuators of the
UAV [69], [70], [71]. It is important to note that the lowlevel controllers (innermost) should have a quicker response
than the higher ones. For a particular case of hybrid UAVs,
the decomposition of the states could result in a low-level
B. Control Laws Classification

Apart from that, flight control systems can be classified
into linear and nonlinear based on the dynamics of the
hybrid UAV model. As previously mentioned, hybrid UAVs
models are nonlinear. However, those models are commonly
linearized using relative equilibrium conditions. Although
811
linear controllers are simple, easy to implement, reduce the

computational effort and minimize the design time but their
performance degrade when operating away from the local
equilibrium point or while performing agile maneuvers. This
is very critical during the transition flight for the case of
hybrid UAVs because changing from vertical flight mode to
horizontal flight mode and vice versa results in operation
far away from the relative equilibrium condition. That is the
reason behind which some current hybrid UAVs implement
nonlinear controllers such as [41], [42], [52], [54], [57],
[68], [73], [75], [76], [79], [81], [82] or three separate linear
controllers, one for the horizontal mode, one for the vertical
mode and one for the transition such as [18], [19], [32], [76],
[82], [83].
Nonlinear controllers are extensively studied and investigated theoretically for application in hybrid UAVs but in
terms of implementation, linear controllers are far more
popular. However, nonlinear controllers operate in a much
wider profile than linear ones which are restricted within a
specific operating region. Also, they consider the full and
true dynamics of the UAV and account for the nonlinear
aerodynamic and kinematic effects, actuator saturations and
rate limitations [84].
The classical PID controller and the Linear Quadratic
Regulator (LQR) are the most common linear control laws
applied in hybrid UAVs while the back-stepping, gainscheduling and dynamic inversion are the common nonlinear
laws. Table II shows a summary of the control laws with their
advantages, disadvantages and applications.
1) Proportional-Integral-Derivative (PID) Controller:
PID controllers are very common in the UAV field. [18],
[19], [32], [35], [43], [44], [45], [52], [53], [55], [58], [72],
[74], [76], [81], [82], [83], [85] have applied PID controllers
in their hybrid UAVs. The PID control law consists of a
proportional, integral and derivative elements. When utilizing
the PID control law algorithm, it is essential to decide
which of these three elements are to be used since each
has particular effect on the control signal as given in [69],
[70], [72]. Generally, for the hybrid UAVs, three main controllers are implemented: Proportional-Integral-Derivative
(PID), Proportional-Integral (PI) and Proportional (P). The
controller gain values are determined by empirical tuning
until some preconceived ideal response of the system is
achieved. Those gains are sometimes estimated theoretically
using Ziegler and Nichols method to reduce the amount of
tuning [69]. It is important to note that the PID controllers
can be implemented in the hybrid UAVs for altitude control,
attitude angles control and velocity control by just changing
the control gains accordingly. Since PID control strategy only
requires appropriate adjustment of the control gains, it serves
as a concrete starting design point for many hybrid UAVs
as it does not require extensive knowledge of the model.
However, PID controller is applicable only for SISO systems,
therefore it does not account for the cross coupling effects
present in UAVs. For such cases, multiple independent PID
controllers are usually utilized in the hybrid UAVs such as
in [35], [43], [52], [72], [55], [73], [76], [80], [82], [85].
2) Linear Quadratic Regulator (LQR) Controller: LQR

controllers goal is to find a control input of the form, that
minimizes the performance index, , which is given by
=
1
2
tf
[xT (t)Qx(t) + uT (t)Ru(t)]dt
(1)
x(t) = Ax(t) + Bu
(2)
t0
subject to
where x(t) Rn is the (n1) state vector, u(t) Rm is

the (m1) control input vector, A is the system matrix, B is
the control influence matrix, and R and Q are real positive
weighting matrices. The feedback control input is of the form
u(t) = R1 (t)BT (t)P(t)x(t)
(3)
where P(t) is known as the Riccati matrix. The only design

freedom in this approach is choosing the weighting matrices.
Brysons Rule is usually used to select those weighting
matrices based on normalizing the signals [70] [88]. Therefore, LQR easily handles complex dynamic systems and
multiple actuators [70]. It is robust with respect to process
uncertainty, asymptotically stable given that the system is
at least controllable and has very large stability margins to
errors in the loop (gain margin of infinity for gain increase
and -6 dB for gain decrease and phase margin of 60 for
each control signal) [70] [89]. On the other hand, LQR
requires access to the full state which is not always possible
[70]. Due to their robustness, LQR controllers are extremely
suitable for hybrid UAV flight control systems and they are
implemented in [56], [57], [14], [76], [86].
3) Backstepping: In Hybrid UAVs, there are several parameters that irregularly change especially during the transition between horizontal and vertical flight modes. Therefore,
a controller that handles these parameters changes might be
necessary. One example is the back-stepping controller which
is based on Lyapunov stability and provides a powerful recursive approach for nonlinear systems that can be transformed
into triangular form. The key idea is to let certain states act
as virtual controls of others. For a system
x
zi
= f0 (x) + g0 (x)z1
= 8 fi (z1 , z2 , . . . , zn ) + gi (z1 , z2 , . . . , zn )zi+1
(4)
(5)
where z represent the general coordinates of the aircraft for

i = 1, . . . , n and zn+1 = u which is the control input. The goal
is to design a feedback control law to stabilize z. The detailed
back-stepping procedure is described in [84], [90], [91]. This
method is beneficial for the hybrid UAVs since it takes into
consideration all the states of the system and accounts for
the nonlinearities present in the model. From the literature
review conducted, it was observed that back-stepping method
was mostly coupled with using Euler-Lagrange approach for
the dynamic modeling as in [42], [54], [68], [79], [87].
812
TABLE II
C ONTROL L AWS C LASSIFICATION OF THE HYBRID UAV S .
Control Law
Advantages
Disadvantages
Applications
PID
Easy implantation, very common control

scheme design in real life applications, does
not require the knowledge of the UAV model
Poor robust ability compared with the robust

controller when the system encounters to multiple challenges, not optimal solution
[18], [19], [32], [35], [43], [44],

[45], [52], [53], [55], [58], [72],
[74], [76], [81], [82], [83], [85]
LQR
Handles complex dynamic systems and multiple actuators, robust w.r.t process uncertainty,
asymptotically stable for controllable systems,
very large stability margins to errors in the loop
Requires access to the full state which is not

always possible
[56], [57], [14] [76], [86]
Backstepping
Very robust for external disturbances and irregular parameter uncertainties, deals with all
the states of the system and accounts for the
nonlinearities
Not optimal, Time inefficient
[42], [54], [68], [79], [87]
GainScheduling
Allows easy understanding and simple implementation of the control laws over the full
flight envelope
Time inefficient
[52], [73], [76], [53], [66], [82]
NDI
Closed loops can be easily tuned
Requires a precise knowledge of the aerodynamic coefficients
[41], [52], [73], [81]
4) Gain-Scheduling: As previously mentioned, the UAV

dynamics could be linearized by applying relative equilibrium conditions around a steady state operating point
and then applying linear controllers such as PID or LQR.
However, the controller performance degrades effectively
when deviating away from that point. Therefore, a prevailing
control design approach, known as gain-scheduling, is to
divide the flight envelope into a finite number of small partitions. For each small region, the UAV dynamics is linearized
around a corresponding steady state operating point. Then,
linear controllers such as PID or LQR each having different
consistent control gain values could be applied for each small
region effectively. Therefore, this allows easy understanding
and simple implementation of the control laws over the full
flight envelope. However, since for each small region, a linear
controller has to be designed, this method might be tedious
and time consuming [70][91]. For the hybrid UAVs, Gainscheduling was mainly utilized to enhance the control during
transition as in [52], [57], [66], [73], [76], [82].
5) Nonlinear Dynamic Inversion (NDI): The dynamic
model of a SISO system can be written in companion form
as
x1
x2
0
.. .. ..
. . .
(6)
=
+
u
xn1 xn 0
xn
b(x)
a(x)
for simplicity, however, the concept could be generalized

for MIMO systems as detailed in [92]. An example of this
approach is given in fig 8 which illustrates how the NDI
linearizes the inner loop making the dashed box a linear
system [91].
Unlike Gain-scheduling, a single controller is required for
the full flight envelope. Another advantage of this method is
that closed loops can be easily tuned as in PID controllers.
However, to apply this method a precise knowledge of the
aerodynamic coefficients is necessary [91]. From the review,
it was noted that NDI was as common and effective as gain
scheduling. It was implemented and tested in many hybrid
UAVs such as in [41], [52], [73], [81].
6) Other Control Strategies: There are several other controlling methods implemented in the hybrid UAVs. [44], [54],
[81] apply adaptive control techniques which account for the
nonlinearities and uncertainties present in the model. J.A.
Guerrero et al. [93] presents a robust control design based
in sliding mode of a mini birotor tail-sitter for the hovering
mode. The work in [94] shows the control of hovering flight
and vertical landing using optical flow. Fault tolerant flight
control system for a tilt-rotor UAV was discussed by S. Park
et al. in [86]. Moreover, other control strategies based on
Lyapunov stability concepts can be found in [46], [58], [59],
[60], [61], [87].
V. C ONCLUSION
In this form, b(x)and a(x) are functions of the state vector

x which linearize the nth state which is the only state affected
by the input. All other elements of the state vector derivative
are linear. In a similar manner to the back-stepping approach,
a virtual control input could be defined as
v = b(x) + a(x)u
A technical overview of the hybrid UAVs has been provided in this paper. The common platform design types and
technical details are introduced first. The modeling is then
explained in terms of model structure, and representative
work on the hybrid UAV dynamics modeling is comprehensively addressed. Representative flight control strategies
implemented in the hybrid UAVs are then discussed and compared in terms of theory, linearity and implementation. The
review presented in this paper is expected to be informative
(7)
which is a linear relation and therefore can be used to

control the system easily. Here, a SISO system is considered
813
to the researchers who are interested in the promising hybrid

UAV development.
[25] A. Vargas-Clara and S. Redkar, Dynamics and control of a stop

rotor unmanned aerial vehicle, International Journal of Electrical and
Computer Engineering (IJECE), vol. 2, no. 5, pp. 597608, 2012.
[26] Introducing the new arcturus jump fixed wing vtol uav. Arcturus UAV.
[Online]. Available: http://www.arcturus-uav.com/aircraft jump.html
[27] R. Park, Arcturus uav upgrades the jump15 vtol uav, Airlines &
Aviation, Aerospace & Defense, December 2014.
[28] R. Tomkins. (2014, May) Airbus vtol quancruiser uav successfully
transitions to fixed-wing flight. Space Daily. [Online]. Available: http://www.spacedaily.com/reports/Airbus VTOL Quancruiser
UAV successfully transitions to fixed-wing flight 999.html
[29] Quadcruiser, an innovative hybrid aircraft concept, Airbus Defence
and Space in cooperation with Airbus Group Innovations.
[30] R. Stone and G. Clarke, The t-wing: a vtol uav for defense and
civilian applications, University of Sydney, 2001.
[31] Z. Omar, C. Bil, and R. Hill, The development of a new vtol uav
configuration for law enforcement, 2008.
[32] M. E. Argyle, R. W. Beard, and S. Morris, The vertical bat tailsitter: Dynamic model and control architecture, in American Control
Conference (ACC), 2013. IEEE, 2013, pp. 806811.
[33] M. Aksugur and G. Inalhan, Design methodology of a hybrid
propulsion driven electric powered miniature tailsitter unmanned aerial
vehicle, Journal of Intelligent and Robotic Systems, vol. 57, no. 1-4,
pp. 505529, 2010.
[34] H. Stone and K. Wong, Preliminary design of a tandem-wing tailsitter uav using multi-disciplinary design optimization, in AUVSIPROCEEDINGS-, 1996, pp. 163178.
[35] H. Menno and C. Notteboom, Design and control of an unmanned
aerial vehicle for autonomous parcel delivery with transition from
vertical take-off to forward flight, Unpublished master dissertation,
KU Leuven, Belgium, 2014.
[36] D. D. Sander Hulsman, Jurjen de Groot, Atmos uav, Leonardo Times,
March 2013.
[37] Interview with the inventors of the quadshot a remote
controlled aircraft. [Online]. Available: http://ideasuploaded.com/
inventor-interviews/
[38] B. Coxworth. (2011, August) Quadshot rc aircraft combines
quadricopter hovering with airplane flight. Online. [Online]. Available:
http://www.gizmag.com/quadshot-hovers-andflies/19449/
[39] K. Z. Ang, J. Cui, T. Pang, K. Li, K. Wang, Y. Ke, and B. M.
Chen, Development of an unmanned tail-sitter with reconfigurable
wings: U-lion, in 11th IEEE International Conference on Control &
Automation (ICCA). IEEE, 2014, pp. 750755.
[40] G. Heredia, A. Duran, and A. Ollero, Modeling and simulation of the
hada reconfigurable uav, Journal of Intelligent & Robotic Systems,
vol. 65, no. 1-4, pp. 115122, 2012.
[41] X. Fang, Q. Lin, Y. Wang, and L. Zheng, Control strategy design for
the transitional mode of tiltrotor uav, in 2012 10th IEEE International
Conference on Industrial Informatics (INDIN). IEEE, 2012, pp. 248
253.
[42] F. Kendoul, I. Fantoni, and R. Lozano, Modeling and control of a
small autonomous aircraft having two tilting rotors, in 2005 and 2005
European Control Conference, 44th IEEE Conference on Decision and
Control, CDC-ECC05. IEEE, 2005, pp. 81448149.
[43] C. Papachristos, K. Alexis, and A. Tzes, Design and experimental
attitude control of an unmanned tilt-rotor aerial vehicle, in 2011 15th
International Conference on Advanced Robotics (ICAR). IEEE, 2011,
pp. 465470.
[44] D. A. Ta, I. Fantoni, and R. Lozano, Modeling and control of a tilt
tri-rotor airplane, in American Control Conference, 2012, pp. 131
136.
[45] C. Papachristos and A. Tzes, Modeling and control simulation of
an unmanned tilt tri-rotor aerial vehicle, in 2012 IEEE International
Conference on Industrial Technology (ICIT). IEEE, 2012, pp. 840
845.
[46] G. Flores and R. Lozano, Transition flight control of the quadtilting rotor convertible mav, in 2013 International Conference on
Unmanned Aircraft Systems (ICUAS). IEEE, 2013, pp. 789794.
[47] C. Abdollahi, Aerodynamic analysis and simulation of a twin-tail
tilt-duct unmanned aerial vehicle, Ph.D. dissertation, 2010.
[48] J. Abras and R. Narducci, Analysis of cfd modeling techniques over
the mv-22 tiltrotor, in American Helicopter Society 66th Annual
Forum, 2010, pp. 1113.
[49] M. Kristi, Stability and control modeling of tilt-rotor aircraft, Ph.D.
dissertation, Maryland: University of Maryland, 2007.
R EFERENCES
[1] B. Defense, V-22 osprey, Space and Security.
[2] D. C. Dugan, Thrust control of vtol aircraft part deux, in the 5th
Decennial AHS Aeromechanics Specialists Conf, January 2014.
[3] B. Handy, Harrier gr7, Royal Air Force Aircraft and Weapons, pp.
89.
[4] U. S. N. C. Newsletter, Rollout week, 2009.
[5] J. Richmond, Its a helicopter! its a plane, Military Aerospace
Technolgy, High Technology, pp. 68 69, 1985.
[6] M. Streetly, IHS Janes all the world aircraft: Unmanned 2013-2014.
IHS, 2013.
[7] Mini panther fixed wing vtol mini uas,. IAI Panther. [Online].
Available: http://www.iai.co.il/2013/35673-41637-en/IAI.aspx
[8] U. Ozdemir, Y. Aktas, A. Vuruskan, Y. Dereli, A. Tarhan,
K. Demirbag, A. Erdem, G. Kalaycioglu, I. Ozkol, and G. Inalhan,
Design of a commercial hybrid vtol uav system, Journal of
Intelligent & Robotic Systems, vol. 74, no. 1-2, pp. 371393, 2014.
[Online]. Available: http://dx.doi.org/10.1007/s10846-013-9900-0
[9] Y. O. Aktas, U. Ozdemir, Y. Dereli, A. F. Tarhan, A. Cetin, A. Vuruskan, B. Yuksek, H. Cengiz, S. Basdemir, M. Ucar et al., A low
cost prototyping approach for design analysis and flight testing of
the turac vtol uav, in 2014 International Conference on Unmanned
Aircraft Systems (ICUAS). IEEE, 2014, pp. 10291039.
[10] A. Vuruskan, B. Yuksek, U. Ozdemir, A. Yukselen, and G. Inalhan,
Dynamic modeling of a fixed-wing vtol uav, in 2014 International
Conference on Unmanned Aircraft Systems (ICUAS). IEEE, 2014,
pp. 483491.
[11] S. Carlson, A hybrid tricopter/flying-wing vtol uav.
American
Institute of Aeronautics and Astronautics.
[12] Firefly6. Birds Eye View. [Online]. Available: http://www.
birdseyeview.aero/products/firefly6
[13] M. Hirschberg, Project zero: The exclusive story of agustawestlands
all electric technology incubator, Vertiflite, vol. 59, no. 3, p. 10=14,
May-June 2013.
[14] O. Tekinalp, T. Unlu, and I. Yavrucuk, Simulation and flight control
of a tilt duct uav, in 2009 AIAA Modeling and Simulation Technologies Conference, Chicago, IL, 2009, pp. 1013.
[15] J. Holsten, T. Ostermann, and D. Moormann, Design and wind tunnel
tests of a tiltwing uav, CEAS Aeronautical Journal, vol. 2, no. 1-4,
pp. 6979, 2011.
[16] T. Ostermann, J. Holsten, Y. Dobrev, and D. Moormann,
Control Concept of a Tiltwing UAV During Low Speed Manoeuvring.
Edinburgh, UK: Optimage Ltd., 2012, 1 CD-ROM. [Online].
Available: http://publications.rwth-aachen.de/record/97342
[17] J. Holsten, T. Ostermann, Y. Dobrev, and D. Moormann, Model
Validation of a Tiltwing UAV in Transition Phase Applying Windtunnel
Investigations. Edinburgh, UK: Optimage Ltd., 2012, 1 CD-ROM.
[Online]. Available: http://publications.rwth-aachen.de/record/116436
[18] E. Cetinsoy, E. Sirimoglu, K. T. Oner, C. Hancer, M. Unel, M. F. Aksit,
I. Kandemir, and K. Gulez, Design and development of a tilt-wing
uav, Turkish Journal of Electrical Engineering & Computer Sciences,
vol. 19, no. 5, pp. 733741, 2011.
[19] K. T. Oner, E. Cetinsoy, M. Unel, M. F. Aksit, I. Kandemir, and
K. Gulez, Dynamic model and control of a new quadrotor unmanned
aerial vehicle with tilt-wing mechanism, 2008.
[20] D. K. Koji Muraoka, Noriaki Okada, Quad tilt wing vtol uav:
Aerodynamic characteristics and prototype flight, in AIAA Infotech@Aerospace Conference. American Institute of Aeronautics
and Astronautics, 2009.
[21] J. Dickeson, D. Miles, O. Cifdaloz, V. Wells, and A. Rodriguez,
Robust lpv h gain-scheduled hover-to-cruise conversion for a tiltwing rotorcraft in the presence of cg variations, in 2007 46th IEEE
Conference on Decision and Control, Dec, pp. 27732778.
[22] High speed tilt-rotoer ousts x-wing project, Flight Internation Technology, January 1988.
[23] J. T. McKenna, One step beyond, rotor wing, February 2007.
[24] A. V. Clara and S. Redkar, Dynamics of a vertical takeoff and
landing (vtol) unmanned aerial vehicle (uav), International Journal
of Engineering Research & Innovation, vol. 3, no. 1, 2011.
814
[50] K. T. Oner, E. Cetinsoy, E. Srmoglu, C. Hancer, T. Ayken, and

M. Unel, Lqr and smc stabilization of a new unmanned aerial
vehicle, 2009.
[51] K. T. Oner, E. Cetinsoy, E. SIRIMOGLU, C. Hancer, M. Unel,
M. F. Aksit, K. Gulez, and I. Kandemir, Mathematical modeling and
vertical flight control of a tilt-wing uav, Turkish Journal of Electrical
Engineering & Computer Sciences, vol. 20, no. 1, pp. 149157, 2012.
[52] Y. Jung, D. H. Shim, and N. Ananthkrishnan, Controller synthesis
and application to hover-to-cruise transition flight of a tail sitter uav,
KAIST, Daejeon, Republic of Korea, pp. 305701, 2010.
[53] P.-R. Bilodeau and F. Wong, Modeling and control of a hovering mini
tail-sitter, International Journal of Micro Air Vehicles, vol. 2, no. 4,
pp. 211220, 2010.
[54] N. B. Knoebel and T. W. McLain, Adaptive quaternion control of
a miniature tailsitter uav, in American Control Conference, 2008.
IEEE, 2008, pp. 23402345.
[55] K. Wong, J. A. Guerrero, D. Lara, and R. Lozano, Attitude stabilization in hover flight of a mini tail-sitter uav with variable pitch
propeller, in IEEE/RSJ International Conference on Intelligent Robots
and Systems (IROS). IEEE, 2007, pp. 26422647.
[56] A. Frank, J. McGrew, M. Valenti, D. Levine, and J. P. How, Hover,
transition, and level flight control design for a single-propeller indoor
airplane. Defense Technical Information Center, 2007.
[57] P. Casau, D. Cabecinhas, and C. Silvestre, Autonomous transition
flight for a vertical take-off and landing aircraft, in 2011 50th
IEEE Conference on Decision and Control and European Control
Conference (CDC-ECC). IEEE, 2011, pp. 39743979.
[58] J. Escareno, S. Salazar-Cruz, and R. Lozano, Attitude stabilization of
a convertible mini birotor, in Computer Aided Control System Design,
2006 IEEE International Conference on Control Applications. IEEE,
2006, pp. 22022206.
[59] A. Sanchez, J. Escareno, O. Garcia, R. Lozano et al., Autonomous
hovering of a noncyclic tiltrotor uav: Modeling, control and implementation, in Proc. of the 17th IFAC Wold Congress, 2008, pp. 803808.
[60] J. Escareno, S. Salazar, and R. Lozano, Modeling and control of a
convertible vtol aircraft, in 45th IEEE Conference on Decision and
Control, San Diego, California. Citeseer, 2006, pp. 1315.
[61] J. Escareno, A. Sanchez, O. Garcia, and R. Lozano, Modeling and
global control of the longitudinal dynamics of a coaxial convertible
mini-uav in hover mode, in Unmanned Aircraft Systems. Springer,
2009, pp. 261273.
[62] J. T. VanderMey, A tilt rotor uav for long endurance operations in
remote environments, Ph.D. dissertation, Massachusetts Institute of
Technology, 2011.
[63] S. Kohno and K. Uchiyama, Design of robust controller of fixedwing uav for transition flight, in 2014 International Conference on
Unmanned Aircraft Systems (ICUAS). IEEE, 2014, pp. 11111116.
[64] P. Casau, D. Cabecinhas, and C. Silvestre, Hybrid control strategy
for the autonomous transition flight of a fixed-wing aircraft, IEEE
Transactions on Control Systems Technology, vol. 21, no. 6, pp. 2194
2211, 2013.
[65] S. R. Osborne, Transitions between hover and level flight for a
tailsitter uav, 2007.
[66] B. L. Stevens and F. L. Lewis, Aircraft control and simulation. John
Wiley & Sons, 2003.
[67] G. Cai, B. M. Chen, and T. H. Lee, Unmanned rotorcraft systems.
Springer Science & Business Media, 2011.
[68] A. Bhanja Chowdhury, A. Kulhare, and G. Raina, Back-stepping
control strategy for stabilization of a tilt-rotor uav, in 2012 24th
Chinese Control and Decision Conference (CCDC). IEEE, 2012,
pp. 34753480.
[69] R. C. Nelson, Flight stability and automatic control. WCB/McGraw
Hill, 1998, vol. 2.
[70] J. P. How, E. Frazzoli, and G. V. Chowdhary, Linear flight control
techniques for unmanned aerial vehicles, Handbook of Unmanned
Aerial Vehicles, pp. 529576, 2015.
[71] R. W. Beard and T. W. McLain, Small unmanned aircraft: Theory and
practice. Princeton University Press, 2012.
[72] A. Manouchehri, H. Hajkarami, and M. Ahmadi, Hovering control
of a ducted fan vtol unmanned aerial vehicle (uav) based on pid
control, in 2011 International Conference on Electrical and Control
Engineering (ICECE), Sept 2011, pp. 59625965.
[73] R. G. N. Ananthkrishnan, H.C. Shim, Controlled near-hover to
cruise transition using a dynamic inversion law, IDeA Research &
Development, American Institute of Aeronautics and Astronautics.
[74] S. Yanguo and W. Huanjin, Design of flight control system for a

small unmanned tilt rotor aircraft, Chinese Journal of Aeronautics,
vol. 22, no. 3, pp. 250256, 2009.
[75] K. Muraoka, N. Okada, D. Kubo, and M. Sato, Transition flight of
quad tilt wing vtol uav, in 28th Congress of the International Council
of the Aeronautical Sciences, 2012, pp. 201211.
[76] R. H. Stone, Control architecture for a tail-sitter unmanned air
vehicle, in 2004. 5th Asian Control Conference, vol. 2. IEEE, 2004,
pp. 736744.
[77] W. S. Levine, The control handbook. CRC press, 1996, ch. 48, pp.
759777.
[78] K. Ogata, Modern Control Engineering, 5th ed. Prentice Hall, 1970,
ch. 8, pp. 567647.
[79] A. Bhanja Chowdhury, A. Kulhare, and G. Raina, A generalized
control method for a tilt-rotor uav stabilization, in 2012 IEEE International Conference on Cyber Technology in Automation, Control,
and Intelligent Systems (CYBER). IEEE, 2012, pp. 309314.
[80] A. Oosedo, A. Konno, T. Matumoto, K. Go, K. Masuko, S. Abiko,
and M. Uchiyama, Design and simulation of a quad rotor tailsitter unmanned aerial vehicle, in 2010 IEEE/SICE International
Symposium on System Integration (SII). IEEE, 2010, pp. 254259.
[81] Y. Jung and D. H. Shim, Development and application of controller
for transition flight of tail-sitter uav, Journal of Intelligent & Robotic
Systems, vol. 65, no. 1-4, pp. 137152, 2012.
[82] T. Ostermann, J. Holsten, Y. Dobrev, and D. Moormann, Control concept of a tiltwing uav during low speed manoeuvring, in Proceeding
of the 28th International Congress of the Aeronautical Sciences: ICAS
Brisbane, Australia, 2012.
[83] J. L. Forshaw, V. J. Lappas, and P. Briggs, Transitional control
architecture and methodology for a twin rotor tailsitter, Journal of
Guidance, Control, and Dynamics, vol. 37, no. 4, pp. 12891298,
2014.
[84] G. Chowdhary, E. Frazzoli, J. P. How, and H. Liu, Nonlinear flight
control techniques for unmanned aerial vehicles.
[85] T. Matsumoto, K. Kita, R. Suzuki, A. Oosedo, K. Go, Y. Hoshino,
A. Konno, and M. Uchiyama, A hovering control strategy for a
tail-sitter vtol uav that increases stability against large disturbance,
in 2010 IEEE International Conference on Robotics and Automation
(ICRA). IEEE, 2010, pp. 5459.
[86] S. Park, J. Bae, Y. Kim, and S. Kim, Fault tolerant flight control
system for the tilt-rotor uav, Journal of the Franklin Institute, vol.
350, no. 9, pp. 25352559, 2013.
[87] G. Flores and R. Lozano, A nonlinear control law for hover to level
flight for the quad tilt-rotor uav, in World Congress, vol. 19, no. 1,
2014, pp. 11 05511 059.
[88] A. E. Bryson, Applied optimal control: optimization, estimation and
control. CRC Press, 1975.
[89] R. Vepa, Flight Dynamics, Simulation, and Control for Rigid and
Flexible Aircraft. CRC Press, 2014, ch. 8, pp. 333486.
[90] A. J. Koshkouei and A. S. Zinober, Adaptive backstepping control of
nonlinear systems with unmatched uncertainty, in Proceedings of the
39th IEEE Conference on Decision and Control, 2000, vol. 5. IEEE,
2000, pp. 47654770.
[91] T. Glad and O. Harkegaard, Flight control design using backstepping, 2000.
[92] M. Karlsson, Control of unmanned aerial vehicles using non-linear
dynamic inversion, Masters thesis, Department of Electrical Engineering, Linkping University, 2002.
[93] J. A. Guerrero, R. Lozano, G. Romero, D. Lara-Alabazares, and
K. Wong, Robust control design based on sliding mode control for
hover flight of a mini tail-sitter unmanned aerial vehicle, in 35th
Annual Conference of IEEE, Industrial Electronics, IECON09. IEEE,
2009, pp. 23422347.
[94] B. Herisse, F.-X. Russotto, T. Hamel, and R. Mahony, Hovering flight
and vertical landing control of a vtol unmanned aerial vehicle using
optical flow, in IEEE/RSJ International Conference on Intelligent
Robots and Systems (IROS). IEEE, 2008, pp. 801806.
815
Renewable and Sustainable Energy Reviews 56 (2016) 12071226
Contents lists available at ScienceDirect
Renewable and Sustainable Energy Reviews

journal homepage: www.elsevier.com/locate/rser
Electric vehicle eet management in smart grids: A review of services,

optimization and control aspects
Junjie Hu a,n, Hugo Morais b, Tiago Sousa c, Morten Lind a
a
Department of Electrical Engineering, Technical University of Denmark, Frederiksborgvej 399, Building 776, room 03, 4000 Roskilde, Denmark
EDF Lab Clamart, Dpartement MIRE, Avenue du Gnral de Gaulle, 92141 Clamart Cedex, France
c
GECAD - Knowledge Engineering and Decision Support Research Center - Polytechnic of Porto (IPP), R. Dr. Antnio Bernardino de Almeida, 431, 4200-072
Porto, Portugal
b
art ic l e i nf o
a b s t r a c t
Article history:
Received 28 April 2015
Received in revised form
30 July 2015
Accepted 11 December 2015
Available online 29 December 2015
Electric vehicles can become integral parts of a smart grid, since they are capable of providing valuable
services to power systems other than just consuming power. On the transmission system level, electric
vehicles are regarded as an important means of balancing the intermittent renewable energy resources
such as wind power. This is because electric vehicles can be used to absorb the energy during the period
of high electricity penetration and feed the electricity back into the grid when the demand is high or
in situations of insufcient electricity generation. However, on the distribution system level, the extra
loads created by the increasing number of electric vehicles may have adverse impacts on grid. These
factors bring new challenges to the power system operators. To coordinate the interests and solve the
conicts, electric vehicle eet operators are proposed both by academics and industries. This paper
presents a review and classication of methods for smart charging (including power to vehicle and
vehicle-to-grid) of electric vehicles for eet operators. The study rstly presents service relationships
between eet operators and other four actors in smart grids; then, modeling of battery dynamics and
driving patterns of electric vehicles, charging and communications standards are introduced; after that,
three control strategies and their commonly used algorithms are described; nally, conclusion and
recommendations are made.
& 2015 Elsevier Ltd. All rights reserved.
Keywords:
Electric vehicles
Fleet operator
Optimization and control strategies
Smart charging
Vehicle-to-grid
Contents
1.
2.
3.
4.
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Service dependent aggregation and its facilitator eet operator. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.
Role of electric vehicle eet operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2.
Service relationships between eet operators and other actors in a smart grid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2.1.
Providing ancillary services to transmission system operator (TSO) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2.2.
Providing ancillary services to distribution system operator (DSO) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2.3.
Providing storage services to renewable energy source supplier (RES). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2.4.
Providing charging cost minimization service to electric vehicle owner . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2.5.
Analysis of the control objectives of eet operator's smart charging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Smart charging infrastructure and modeling of EV battery dynamics and driving pattern . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.1.
Battery modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2.
Charging standards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.3.
Communications standards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.4.
Driving pattern . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Control strategies of eet operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.1.
Centralized control of electric vehicle eet operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Corresponding author. Tel.: 45 93511212; fax: 45 45886111.

E-mail addresses: junhu@elektro.dtu.dk (J. Hu), hugo.morais@edf.fr (H. Morais), tabsa@isep.ipp.pt (T. Sousa), mli@elektro.dtu.dk (M. Lind).
http://dx.doi.org/10.1016/j.rser.2015.12.014
1364-0321/& 2015 Elsevier Ltd. All rights reserved.
1208
1209
1209
1209
1209
1210
1211
1212
1212
1213
1213
1213
1213
1214
1214
1214
1208
J. Hu et al. / Renewable and Sustainable Energy Reviews 56 (2016) 12071226
4.2.
Transactive control of electric vehicle eet operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3.
Price control of electric vehicle eet operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.4.
Discussion on integrating the control strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5. Mathematical modeling and control algorithms: centralized control of eet operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.1.
Linear programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.2.
Quadratic programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.3.
Dynamic programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.4.
Mixed-integer linear programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.5.
Mixed-integer non-linear programming. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.6.
Stochastic programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.7.
Robust optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.8.
Heuristic and meta-heuristic algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.9.
Model predictive control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.10. A summary of the presented algorithms with three types of criteria. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6. Mathematical modeling and control algorithms: transactive control of eet operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7. Mathematical modeling and control algorithms: price control of eet operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8. Conclusion and recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.1.
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.2.
Recommendations on future research directions in the area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1. Introduction
To achieve the European energy roadmap 2050 [1] The EU is
committed to reducing greenhouse gas (GHG) emissions to 80
95% below 1990 levels by 2050 in the context of necessary
reductions by developed countries as a group. Carbon emission
reduction target are also set up in other countries such as China,
United States, and Korea. The corresponding inuencing factors
studies [26] show that power industry and transportations are
the major contributing factors to CO2 emission. For example, the
power industry was responsible for around 30% of GHG in EU-27
in 2011. The second sector with more GHG emissions in 2011 were
the transports with a share of 20.3% [7]. Thus, the decarbonization
of two main activities including power industry and transportations is required.
To reduce the GHG emissions, in power systems, renewable
sources such as wind and solar power are widely adopted [8,9]. In
the transport sector, electric vehicles (EVs) are important means to
assure the GHG emission reduction goals [4,10]. However, the
increasing investments in renewable power bring operational
challenges into the power systems due to the intermittent
resources. To cope with the new challenges, in smart grid [11,12],
EVs are commonly recognized as one solution in addition to their
environmental benets in the transport sector. They can be utilized to balance power uctuations caused by the high penetration
of intermittent renewable energy sources [13]. However, a largescale integration of EVs also means new loads to electric utilities,
and undesirable congestions and voltage problems may exist in
the distribution network during the charging process [14]. All
these factors bring new challenges to power system operators. As
an outcome, smart charging solutions (including power to vehicle
and vehicle-to-grid (V2G)) are needed in order to make EVs an
asset to the smart grid rather than a merely traditional load.
Much research has been done to capture the benets of electric
vehicles as well as to address the conicting challenges. It is
concluded from the literature that a new business entity, namely
an EV eet operator (FO) has been widely proposed to exploit the
new business opportunities by providing the multiple services of
EVs to system operators. Alternative names for an EV FO are used
such as EV virtual power plant, EV aggregator, EV charging service
provider or EV service provider. The new entities [15,16] could be
independent or integrated in an existing business function of the
energy supplier or distribution system operator (DSO).
1214
1215
1215
1216
1216
1216
1217
1217
1217
1218
1218
1219
1219
1219
1221
1222
1223
1223
1224
1224
In principle, three types of control strategies can be used by FOs

when aiming at the objectives mentioned above, namely centralized control, decentralized control considering the distinctive
market-based/transactive control and price control, respectively.
Centralized control means that FOs directly schedule and control
the charging of electric vehicles, while the transactive control and
price control are usually implemented in a form of price signal, i.e.
the individual EV optimizes the charging based on the electricity
price information made available to them from FOs or from DSO.
The key difference between transactive control and price control is
the information exchange requirements, i.e., transactive control
requires explicit responds from the individual EV while price
control does not need such responds. More discussions regarding
these three control methods will be presented in Section 4. From
the discussions [17,18], it is generally accepted that for the centralized control the decisions are made on the system-level and
therefore it can give better results such as ensuring the security of
the power systems; however, the cost of communication infrastructure would be high for centralized charging. For the transactive control and price control, one of main advantages is the
possibility to minimize the communications infrastructure cost
[19], nevertheless, the solution may or may not be optimal,
depending on the information sharing and the methods used to
make the charging schedule.
The main content of this paper is to give a review and classication of the optimization and control strategies used for smart
charging of EV eets. Although the authors are aware of similar
works [2025] on reviewing smart charging of electric vehicles
that are published recently, the research contributions of this
study include: 1) a comprehensive summary of service relationships between eet operators and other four actors in a smart grid
context are presented; 2) three control strategies used by FOs
including centralized control, transactive control, and price control
are specically distinguished and discussed; 3) mathematical
modeling methods of the three control strategies are compared
and evaluated. The aim of the present study is to provide a comprehensive understanding about EV eet management that allows
commercial actors, e.g., EV FOs to exploit the service based electric
vehicle aggregation and make the electric vehicles become integral parts of smart grids.
The remainder of the paper is organized as follows: the role
and the service dependent aggregation of electric vehicle eet
operator are discussed in Section 2. Section 3 introduces the
modeling of EV battery and driving pattern, charging and communication standards. Three control strategies including centralized control, transactive control and price control are described
in Section 4. Commonly used algorithms in the centralized control,
transactive control and price control of smart charging of EVs are
presented in Sections 57, respectively. Section 8 concludes the
paper with some suggestions for future research.
2. Service dependent aggregation and its facilitator eet

operator
In [26], Lopes et al. shortly summarized that a large deployment of EVs will involve the following studies: 1) evaluation of the
impacts that battery charging may have on power system operation; 2) identication of adequate operational management and
control strategies regarding batteries' charging periods; 3) identication of the best strategies to be adopted in order to use preferentially renewable energy sources (RES) to charge EVs; 4)
assessment of the EV potential to participate in the provision of
power system services, including reserves provision and power
delivery. Following the summary in [26], this study starts by
reviewing four kinds of control objectives of operational management of an EV eet. Furthermore, it is possible to see these four
control objectives as four types of services that can be offered by
FOs to other actors in smart grids. In this section, the role of EV
eet operator is rstly discussed; then the relation between the
eet operators and other actors in a smart grid are described; next,
four kinds of services which can be provided by eet operators are
introduced; at the end of this section, the relationship between
these four kinds of services is analyzed.
2.1. Role of electric vehicle eet operators
San Romn et al. [16] proposed a regulatory framework for
charging EVs where two electricity market agents, an EV charging
manager and an EV aggregator/eet operator are introduced. The
EV charging manager is responsible to develop the charging
infrastructure. The EV eet operator is responsible for providing
charging services to EV eet and managing the EV eet for other
services provision. With respect to the feasibility of applying the
eet operator concept, Bessa and Matos [15] gave a literature
review regarding the economic and technical management of an
EV aggregation agent. The reviewed paper [15] is organized into
three subject categories: electricity market, EV technical and
economic issues; aggregation agent concept, role and business
model, and algorithms for EV management as a load/resource.
It is observed that the role of each type of FO proposed various
and the main difference lies in whether the FO has two functions
or one function, i.e., some studies assumed that a FO is both a
charging equipment supplier and charging service provider, others
only refer eet operator as the charging service provider. Although
several differences exist in the details of the proposed FO concepts,
they are assumed to achieve the same goals in this study,
regardless the ownership of the charging equipment. These goals
are:
Guarantee driving needs of the EV owners with optimal management of EV charging;
Provide peak power to the electric network from the V2G

capability;
Provide ancillary services to power system operators with

optimal allocation of EV eet resources.
1209
2.2. Service relationships between eet operators and other actors in

a smart grid
Fig. 1 illustrates the relationships between eet operators and
other actors in a smart grid by showing the four services that FOs
can provide to them.
Note that the relationship between FOs and EVs is slightly more
complex. From one perspective, FOs need to ensure the participation of EVs and then have the capability of providing services to
other actors in the smart grid; from another perspective, FOs can
provide service of energy trading to EVs such as helping the EV
owner to save money. Therefore, FO may need to consider more
factors rather than purely make prots when providing services
to EVs.
2.2.1. Providing ancillary services to transmission system operator
(TSO)
Ancillary services are needed to maintain the balance between
the supply and demand so that a secure and reliable functioning of
all power system is ensured [27]. The regulation down, regulation
up, spinning reserve and non-spinning reserve are the most
common services in the frequency control [27]. Regulation down
and up, also referred as automatic generation control, has the
purpose to ne-tune the frequency by matching supply and
demand at any time that it is needed to respond within a minute
or less. More specically, regulation down and up are characterized by decreasing and increasing, respectively, the actual operation power level of a specic resource that participates in this
service [27]. Spinning reserve corresponds to unloaded synchronized generation capacity that can quickly provide power to the
network [27]. Non-spinning reserve is the portion of generation
capacity that is capable of being synchronized and ramping to a
specied load [27]. The regulation (up and down) is the service
with the quickest response, the shortest duration, the shortest
service availability and the highest price of the frequency regulation services. In terms of hierarchy for power system operators
these services are categorized by: 1) regulation up and down; 2)
spinning reserve; 3) non-spinning reserve. For this reason the
regulation market contains the highest prices of all frequency
service markets.
The operation of power system will benet greatly from the
introduction of EVs participation, namely in the electricity markets
related to ancillary services. Moreover, the large integration of
intermittent renewable sources will have more ancillary services
requirements to maintain the power system balanced. Kempton
and Tomic [28,29] analyzed the potential of EVs in the ancillary
service market and concluded that the participation in the regulation market appears to be one of the most promising application, because it can offer a substantial earning potential to EV
owners. Dallinger et al. [30] proved the effectiveness to apply EVs
in the frequency regulation and they could make an attractive
alternative to the large generators with high prices. Rotering and
Ilic [31] studied the participation of V2G in the ancillary service
markets for the independent system operator of California. Based
on this study, provision of regulating power substantially improves
electric vehicle economics and the daily prots amount to 1.71$.
Almeida et al. [32] proposes a novel primary frequency control
technique for isolated systems considering EVs penetration. Sortomme and El-Sharkawi [33] analyses the potential of V2G to
participate in the regulation and spinning reserve based on the
Electric Reliability Council of Texas market. The authors concluded
that EVs can bring great benets to the regulation and spinning
reserve, but the battery life cycle represents a challenge to the
viability for the application of V2G mode to the ancillary service
market.
1210
Fig. 1. The services relationships between eet operator and other actors in a smart grid.
Table 1
Characterization of the literature by the frequency regulation market.
Service name
Short description
Expected asset in the market (e.g. Danish and European) References
Regulation up and down A reserve that stabilizes the frequency in the matter of seconds of
minutes
Spinning reserve
Release the primary reserve and restore the frequency to 50 Hz
Non-spinning reserve
A manual reserve that releases the other two reserves
Overall, the above studies have pointed out that EVs can have a
high contribution to the regulation and spinning reserve services,
especially in the regulation up and down services. These services
require a quick response from the resources (e.g. generators,
demand response programs and EVs) that can be called for a few
minutes. In addition, an adequate telecommunication infrastructure is required to support the ancillary services responding
from the resources. Since EV batteries have characteristic of rapid
responses, EVs are most appropriate for the regulation up and
down. Additionally, the highest price offered in regulation up and
down services can bring more prots for the EV owners. Regarding
spinning reserve, it is also mentioned as a good service for EVs
participation, but with lower revenue, higher duration and higher
availability than regulation service. The spinning service requires
that resources can have a synchronized reserve available to be
called in a typical duration of minutes (e.g. 10 min) until 1 h. With
respect to the non-spinning reserve it is pointed out by the literature that EVs will not be suitable for this service due to its
higher duration (e.g. hours) and lower prices. However, accelerated battery degradation might be the main challenge for a complete implementation of EVs participation in the ancillary services.
Table 1 gives an overview of the literature by the different kind of
frequency regulation services (regulation, spinning and nonspinning reserves).
Yes
[2833]
Yes
No
[28,29,33]
N.A.
2.2.2. Providing ancillary services to distribution system operator

(DSO)
Research indicates that uncontrollable charging of a large scale
of EVs will bring some challenges to the distribution network. The
challenges are related to peak power problem, grid congestion,
power losses and voltage drop. Several studies evaluate the performance of the distribution network considering three types of
strategies [20,23]:
Uncoordinated charging;
Smart charging;
Smart charging and discharging (i.e. bidirectional V2G).
An uncoordinated charging occurs when EVs connect to the
distribution network and start immediately to charge until the
batteries reach their capacity. The smart charging consists in an
actor (e.g. DSO or EVs eet operator) controlling the time and
power of the EVs charging. The V2G control is similar to the smart
charging and it has the capability of providing power to the grid.
Studies concerning the charging rate control of EVs and their
effect on the distribution network are dated back to the early
1980s [34]. Heydt [34] argued that load management should be
deployed to ease peak loading by the DSO, which is measured in
term of load factor improvement. In 1993, Rahman and Shrestha
[35] indicated that even low penetration levels of EVs can create
new peak loads exceeding the natural peak if sufcient attention is

not paid to distribute the charging load throughout the off-peak
periods. A penetration level of 20% is found to be the upper limit
which could be managed by distributing the charging load. The
literature points that without smart charging the power consumption on a local scale can lead to grid problems, such as
unpleasant load peaks, line congestion and voltage limit violation.
The studies [34,35] examined mainly the impacts of EVs to the
distribution system by adding the EV loading prole to the already
existing load prole and evaluating the overall effect.
Recently, more parameters such as power losses, load levelling
and grid congestion have been considered for supporting the
integration of EVs in the distribution network. Sortomme et al.
[36] proposed optimal charging algorithms to minimize the
impacts of EVs charging in the distribution network in terms of
power losses and load factor. Morais et al. [37] proposed a multiobjective optimization algorithm to evaluate the impact of EVs in
the power demand considering the scheduling cost and load
levelling as objectives functions of the proposed algorithm. In
terms of voltage regulation, this objective is achieved by proper
scheduling of the reactive power or by controlling the load
demand in order to reduce the voltage drop. Wu et al. [38]
examined the potential of a proper selection of current phase
angle by EV charger in order to compensate capacitive and
inductive reactive power. Leemput et al. [39] discussed the impact
of the reactive support by EVs charging in a low-voltage residential distribution network. In [40], a methodology is proposed to
deal with active and reactive management of a distribution network with EVs to improve the voltage prole while a minimum
operation cost is achieved.
Most of the literature indicated that TSO has the responsibility
with the ancillary services about frequency control (i.e. regulation,
spinning and non-spinning reserve), DSO is responsible for the
local voltage control and local congestion prevention. However,
with the increasingly penetration of distributed resources, the DSO
can have an important role in the frequency control in the future,
especially in the regulation and spinning reserve [41]. In [42], it is
presented the application of distributed generation to provide
ancillary services to control the frequency. The FO could also
participate in this service that the DSO will need in the future
operation of the power system.
Currently, the ancillary services proposed in the literature that
FO can provide to DSO are: 1) congestion prevention (i.e. reduce
peak load and power losses); 2) voltage regulation. The literature
is divided based on this classication as it is shown in Table 2.
Nevertheless, in the future the FO can also be able to provide
regulation for the frequency control under DSO supervision.
2.2.3. Providing storage services to renewable energy source supplier
(RES)
One of the main challenges for operating the power system
with renewables sources such as wind and solar is related to their
intermittent behavior that is inuenced by the stochastic nature of
their primary energy sources. Regarding this subject, EVs have
been suggested as one of the most promising solutions for mitigating this intermittent behavior, compared with other solutions
1211
such as using centralized storage system or backup generation that

represent a high capital cost to the power system operators.
Basically, EVs can store the excess energy from renewables in their
batteries that would otherwise be curtailed and wasted. EVs can
use this stored energy for their daily driving, or in the case of EVs
with V2G can also supply this energy back to the network. The
literature concerning this topic is focused more on the backup of
EVs to the wind than the solar energy [22,24,43].
Lund and Kempton [44] investigated the impact of using V2G
technology to integrate the sustainable energy system. Two
national energy systems are modeled; one for Denmark including
combined heat and power (CHP), the other is a similarly sized
country without CHP. The model (EnergyPLAN) integrates energy
for electricity, transport and heat, includes hourly uctuations in
human needs and the environment (wind resource and weatherdriven need for heat). The results indicated that adding EVs and
V2G to these national energy systems allows integration of much
higher levels of wind electricity, and also greatly reduces national
CO2 emissions. Bellekom et al. [45] investigated the impact of large
scale EVs and wind integration in the Dutch power system. The
study concludes that wind integration can increase from 4 GW (no
EVs scenario) to 10 GW if there are around 1 million EVs connected to the network. Dallinger and Wietschel [46] examined the
impact of EVs charging in the German electricity system with 50%
share of renewables in 2030 (wind and solar) where the charging
strategy is obtained through consumer price response. Furthermore, this study concluded that EVs can play an important role in
mitigating the intermittent behavior of renewables, being stored
in the EVs over more than 50% of the yearly excess renewable
energy.
Saber and Venayagamoorthy [47] proposed a particle swarm
optimization algorithm to handle the unit commitment of a power
system considering a large penetration of renewables and EVs. The
authors concluded that using the EVs with V2G in a smart grid
concept will contribute to the minimization of cost and emissions
in the unit commitment and to reduce the unbalance introduced
in the operation of the intermittent renewables. Another study by
the same authors [48] proposes a particle swarm optimization
algorithm to solve the same optimization problem including
uncertainties related to load consumption, renewables generation
power and number of EVs connected to the network. A probabilistic distribution is considered for each stochastic variable and
then a few scenarios are generated and the optimization method
solves the unit commitment for each different scenario. In addition, a mixed-integer linear programming algorithm is presented
in [49] to deal with the resource scheduling in smart grid context
considering a scenario with intensive penetration of renewables
and EVs.
In terms of EVs integration with solar energy, Birnie [50] proposes the installation of photovoltaic (PV) panels to supply the
daytime charging of EVs in a parking lot assuming New Jersey
solar irradiation. The study concluded that the PV panels can meet
the driving needs of the EV owners during the summer, but not in
the winter. Zhang et al. [51] examined the integration of PV panel
in collaboration with EVs and heat pumps in the Kansai Area of
Japan. The authors concluded that it would be necessary to
Table 2
Characterization of the literature by the ancillary services in the DSO.
Service name
Short description
Congestion prevention Intelligent control of the charge and discharge of EVs in order to avoid grid congestion,
minimize load peaks and reduce power losses.
Voltage regulation
Intelligent control of the charge and discharge of EVs to improve the voltage prole
avoiding the increase of voltage drops.
Expected asset in the market (e.g. Danish

and European)
References
Yes
[3437]
Yes
[26,3840]
1212
Table 3
Characterization of the literature by the RES service.
Service name
Short description
Expected asset in the market (e.g. Danish and

European)
Renewable integration Evaluate the share of renewables that is possible to have in the power system with Yes
a large scale of EVs.
Storage device
Large mobile storage capacity to mitigate the intermittent behavior of renewables. Yes
introduce 1 million EVs and 1 million heat pumps to reduce the

excess of energy by 30% from a hypothetical scenario of 30 GW of
solar capacity in the area.
Overall, the literature examines the extent of renewables
integration, namely wind and PV energy, which EVs can accommodate in the power system in order to reduce CO2 emissions. In
general, these studies concluded that the connection of more EVs
in the network and the control of their charging rate have the
potential to increase the share of renewables in the power system
and to mitigate the intermittence behavior of renewables. Besides,
EVs with V2G can supply the excess of renewable energy previous
stored in the batteries back to the network when power is needed,
such as in periods of high demand and low renewable generation.
Table 3 presents a classication of the literature related to this
subject, in which the articles have been divided in studies related
to renewable integration and storage device.
2.2.4. Providing charging cost minimization service to electric vehicle
owner
In the above sections, research has been presented concerning
the impact of EVs on other actors, namely TSO, DSO and RES
supplier. This subsection presents literature considering EV's
impact on the FO and EV owners. In order to stimulate active
participation in smart charging management and to reduce the
initial investments with the purchase of EVs, it is noted that the
proposed works related to optimization and control of charging
and discharging of electric vehicles should be aware of the fact of
sharing prots from the FO with the EVs owners. It is assumed in
[52] that the EV eet operator manages the electricity market
participation of an EV eet and presents a framework for optimal
charging of the EVs. The result illustrated that the electricity bills
of charging the EVs are reduced. In addition, the electricity price of
the day-ahead spot market, the regulation market and the driving
patterns of the EV eet are usually assumed to be known by the
eet operator, who is assumed to be the price-taker in the electricity market. However, Kristoffersen et al. [53] also investigated
the possibilities of EV management where the FO has a signicant
market share and can affect electricity prices by changing the load
through charging and discharging.
Besides studying the optimal charging from an EV eet perspective, research in [31] shows how dynamic programming can
be used by the individual EV controller to make an optimal charging schedule taking into account the electricity market price. In
[33,54], a strategy is presented for an EV aggregator to participate
in the spot and regulation market. The V2G service can bring
benets to the EV owners because it can reduce the cost that
owners had with the charging of their vehicles. However, the literature point out major concerns for the V2G concept [21]: 1)
additional investments for enabling the bidirectional power ow,
2) advanced communication and smart metering and 3) high
degradation of battery because of repeated cycling in comparison
with the scenario of smart charging. It is expected in the near
future, more FOs can exist in the market that provide services to
the EVs to control and optimize their charging. Additionally, it is
also expected that the concerns about V2G can be solved enabling
the FO to offer V2G services for the electricity markets. Table 4
References
[44,45,50,51]
[4649]
Table 4
Characterization of the literature by the EV owner service.
Service name
Short description
Expected asset in
the market (e.g.
Danish and
European)
References
Charging
management
Charging and discharging
management
Charging management
aiming at lower cost
Charging and discharging management aiming at maximum prot
Yes
[31,52,53]
Yes
[33,54]
presents a classication of the literature related to this subject, in

which the articles have been divided in studies related to charging
management and charging and discharging management.
2.2.5. Analysis of the control objectives of eet operator's smart
charging
Several questions naturally arises after reviewing the four main
services (described in 2.2.12.2.4) that FO can provide to other
actors, e.g., whether conicting interests exist between different
services, whether some services can be integrated, and what are
the relationships between these four main services. It is observed
that multi-services research is already performed in several studies, however a systematic way of understanding the relationships
between the described services is missing. Johannsson et al. in [55]
presented a scheme to deal with these relationships based on a
prioritized list. The potential conicts between different actors (i.e.
FO, TSO and DSO) appear when two actors need to use EVs for
conicting services, such as a service activated by an actor can
cause a negative inuence to other actors management.
The authors in [55] present two examples concerning conicting interests between actors. One of these examples is about
how the frequency control can be in conict with peak-shaving. In
the example, the TSO control the frequency by activating a contracted aggregator. Then the aggregator responds by increasing the
aggregated consumption of its consumers, e.g., EVs. However, the
aggregator's positive response can cause overload in local areas,
and the DSO must execute a service of congestion prevention to
control this overload. If the aggregator attends the peak-shaving
incident, the TSO may require another solution to maintain the
frequency under control. However, if the aggregator ignores the
peak-shaving alert, the local area that is managed by the DSO
would be overloaded and more equipment would be damaged,
such as power transformers and lines [55].
The authors in [55] proposed a prioritized service list for
handling potential conicts. The list includes: 1) emergency
actions (TSO); 2) alert actions (TSO/DSO); 3) local voltage control
(DSO); 4) peak-shaving (DSO); 5) voltage support (TSO); 6) Mvar
bands (DSO); 7) frequency control; 8) other ancillary services
(TSO); 9) imbalance issues (Aggregator, e.g. FO); 10) power quality.
This list should be used for helping different actors' management
when causing a negative inuence to other actors. In a case of two
conicting services, the main purpose is a service with higher
priority will be activated instead of the other one. In order to
incorporate the list in the management of the actors, the authors
presented the behavior description that denes the behavior of a

given resource for a specic service.
3. Smart charging infrastructure and modeling of EV battery

dynamics and driving pattern
When designing the control strategies that aim at providing
different services described in the previous section, many aspects
should be taken in consideration namely the battery modeling, the
charging and communication standards and nally the charging
requirements of the users according to their driving proles. In
this section a general overview of these four aspects is presented.
3.1. Battery modeling
The increase of efciency of batteries is one of the most
important challenges in the EVs industry. Several battery technologies including Pb-acid battery, Ni CD battery, Ni-MH battery,
Li-ion battery, and Li-polymer battery are available in the market,
each one with different characteristics meaning different advantages and disadvantages. Three main characteristics of batteries
including the energy efciency, the energy density and the power
density are studies in [56].
In general, there are two ways of modeling the charging characteristics of an EV battery. One is a model for an individual battery pack, another is aggregated or cell based model. For simplicity, most of the studies considered EV as a battery pack when
investigating the optimal charging and discharging problem. Currently, most battery-modeling research [57,58] focus on three
types of studies:
The rst and most commonly used model is termed as a per-
formance or a charge model and focuses on modeling the state

of charge of the battery, which is the single most important
property in system assessments;
The second type of model is the voltage model, which is
employed to model the terminal voltage, so that it can be used
in more detailed modeling of the battery management system
and in more detailed calculation of the losses in the battery;
The third type of model is the lifetime model used for assessing
the impact of a particular operating scheme on the expected
lifetime of the battery.
3.2. Charging standards

In the last decade, standards related to EVs were proposed from
different principles. In general, IEC 62196 are adopted in Europe.
The IEC 62196-1 is based on IEC 61851 dening four charging
modes:
1. Mode 1: AC slow charging from a household socket-outlet.
Mode 1 charging is the most common option for electric
vehicles due to the use of traditional house/industrial socketoutlet [56,59]. Mode 1 charging is now only considered as the
main mode for small vehicles such as two wheelers [59].
2. Mode 2: slow charging from a household-type socket-outlet
with an in-cable protection device in AC. Mode 2 also allows the
use of traditional house/industrial socket-outlet. However, this
charging mode, provides additional protection by adding an incable control box with a control pilot conductor between the
electric vehicle and the plug or control box [56,59];
3. Mode 3: slow or fast charging using a specic EV socket-outlet
with control and protection function installed in AC. Mode
3 requires a dedicated connection between the EVs and the
charging station [56,59];
1213
4. Mode 4: fast charging using an external charger in DC. Two submodes of operation are considered for this mode, namely, the
DC level 1 (voltage lower than 500 V, current lower than 80 A,
power at 40 kW) and the DC Level 2 (voltage lower than 500 V,
current lower than 200 A, power at 100 kW) [56,59].
It is important to mention that in charging Mode 1, there is no
communication between the EVs and the charging point through
the connection system. In Mode 2 and 3, a control pilot communication can be included allowing the control of the charging rate.
In charging Mode 4, it is mainly used for the fast charging. A
communication system is included in charging Mode 4 that allows
battery charging management. In addition, in Mode 2, 3, and 4,
wireless communication systems can be used to communicate
with electric vehicles and control the charge and discharge
process.
Concerning the type of connectors, the IEC 62196-2 proposes
different types based on other standards the SAE J1772-2009 (Type
1) or the VDE-AR-E 2623-2-2 (Type 2). The type 3 is also xed, but
is not yet completely dened. Concerning the fast charging, the
62196-3 denes the DC fast charging plugs to be used in Mode 4,
namely the CHAdeMO and the combined charging systems.
In U.S. the charging standards are dened by the Society of
Automotive Engineers (SAE) in SAEJ1772. The recognized types of
plug are similar to IEC 62196, but SAE has selected the J1772
combo plug as the standard. This plug allows both AC and DC
charging using the same plug. The SAEJ17722011 denes six
charging levels, three in AC and three in DC [21]. The AC Level 1 is
practically applied at home environments while the AC Level 2 is
suitable for public and commercial areas. The DC-fast charging
levels (DC Level 13) are more adapted to public use.
In China, the EVs charging standards are dened in GB/T
202342011. According to this standard, EVs charging can be made
in AC using the Mode 2 and 3 dened in IEC 62196. However, this
use is limited to a single phase system allowing a maximum current of 16 A. The DC fast charging is also available considering a
specic connector that allows a maximum voltage of 750V and
125/250A [60]. The fast charging uses the CATARC (China Automotive Technology and Research Center) protocol. In Japan, the
CHAdeMO specications are used as a standard [61].
3.3. Communications standards
In order to illustrate how the charging and discharging schedule is implemented, this subsection discusses the relevant communication standards for integrating EVs into power grids. It is
noted that the purpose of this section is to provide the relevant/
widely used communication standard that can support EV smart
charging rather than comparing the various communication
standards. The IEC standards illustrated in Fig. 2 is recommended
in [62]. The objectives of the study [62] is to realize a standardized
communication interface between the vehicle and the grid. The
standardization will make it possible for EV users to have easy
access to EV charging equipment (EVSE) and related services
throughout Europe. EVSE refers to all devices installed for delivering power from the electrical supply point to the EV and this
charging equipment will support smart charging functions. The
decision of the charging can be made on the EV level or on the FOs
level. The IEC 15118 is the most recommended communication
standard [62], and is demonstrated in details in [63], by the
sequence diagram of a charging process between the EVSE and
the EVs.
The IEC 61850 is recommended for the communication
between the EVSE and the FOs [63]. In [64], the use of the smart
metering infrastructure to transmit the information concerning
the EVs charging state using the multiple access control protocol is
1214
Fig. 2. Relevant ICT standards support the EV smart charging in the context of smart grids.
proposed. According to [65], the protocol SAE J2836/3TM, published in January 2013, allows the coordination of both distributed
energy resources and electric vehicles including the V2G capability
(control the charge and discharge process). The information
exchange with the EV was derived from the IEC/TR 61850-90-8
from February 2013. In [66], two other promising protocols to be
used in the communication between charging stations and the
system operators and/or aggregators, namely Open Charge Point
Protocol (OCPP) and Hubject's Open Intercharge Protocol (OICP)
are analyzed.
3.4. Driving pattern

The modeling of driving pattern can be divided into two main
aspects:
Use of EVs, in other words, a typical users daily driving

activities;
Location of EVs when charging and how many of them will be

charged at a time.
Kristoffersen et al. [53] investigated a method to construct
driving patterns from the historic data in Denmark. By clustering
survey data of the vehicle eet in Western Denmark (January
2006December 2007), a representative driving pattern for each
vehicle user are constructed. Shahidinejad et al. [67] developed a
daily duty cycle which provides a complete data set for optimization of energy requirements of users. This information can also
be used to analyze the impact of EVs' daytime charging on the
electric utility grid, which may create a peak demand. Normally,
intra city or short-term driving patterns are largely predictable
due to xed working hours and xed business schedules and
routes of EV owners. In [68], a simulation tool is proposed
allowing the generation of driving proles for a large number of
EVs considering a set of probabilities and EVs characteristics like
the percentage of vehicles in movement, batteries efciency, trips
distances, vehicles class distribution, etc.
A more detailed analysis of driving patterns is performed in
[69], considering that the environmental, economic and technical
factors which inuences the driving patterns of EV owners.
Technical factors include: the number of EVs being charged (EV
penetration trend), the availability of charging infrastructure,
charging voltage and current levels, charging time, battery technology, battery life time and capacity. An analysis of real EVs
driving behaviors is performed in [70]. The results indicated that
the use of EVs has changed the daily routines of 36% of the participants leading to a signicant reduction both in energy consumption and in greenhouse gas emissions.
4. Control strategies of eet operator

In this section, the focus is the control strategies of eet
operators, i.e., how the FO optimally schedules and controls the
EVs according to the specied objectives, such as the aforementioned four kinds of services (described in Sections 2.2.12.2.4).
Three control strategies are presented in this section: centralized
control, transactive control and price control. Centralized control
means that FOs directly schedule and control the charging of
electric vehicles [52]. Transactive control [71] is a form of marketbased control method that has been adopted by the GridWise
Architecture Council [72]. The purpose of transactive control is to
reach equilibriums by using a scalable, distributed mechanism via
exchanging information concerning generation, loads, constraints
and responsive assets over dynamic, real-time forecasting periods
using economic incentive signal. PowerMatcher1 [73] is a good
example of using transactive control for supply and demand
matching in electricity networks. Transactive control usually
requires two way communications, e.g., exchange of the price and
power schedule information. Price control [74,75], instead uses
one way communication and applies broadcasting of price signal
with a regular updated frequency to the demand side resources.
An overview and comparison of the three control methods is
presented in Fig. 3.
4.1. Centralized control of electric vehicle eet operator

In this control strategy, the FO will centralize all the relevant
information from the aggregated EVs, as it is shown in Fig. 4. The
FO will require four inputs, i.e. the model of the EV battery and the
EV driving patterns, the grid constraints and the electricity price to
make a proper charging control of EVs.
4.2. Transactive control of electric vehicle eet operator

The information ow in transactive charging control is presented in Fig. 5. Note that the goal of the FO is implicit in the
gure. The key point of this gure is to show the two way information exchanging in term of power schedule and price. The basic
idea of transactive control application in EV charging control is
that EVs update their charging proles independently given the
price signal from the FO until equilibriums are achieved. In the
transactive control, the EVs charging schedule is a result of the
information exchanging between FO and EVs and thus it is not a
purely decision of the EV owner.
1
http://www.powermatcher.net.
Centralized
control
Transactive
control
Price control
Control signals (i.e., set

points)
High level controller
makes the decision
High level controller generates

and sends the price to the units
Low level units respond with
power schedule
Multiple iterations
Privacy improved
High level controller generates

and sends the price to the units
Low level units need not to
explicitly respond.
Privacy improved
Low for Control

object
High for controller
Relative high for

Control object
Low for controller
Low for Control

object
Relative high for
controller
Computational
requirements
Features
1215
One/Two way
High
Two way
Relative high
One way
Low
Communication
form and cost
Fig. 3. Overview of control strategies.
Fig. 4. Primary inputs and output of EV FO in a centralized control strategy.
4.3. Price control of electric vehicle eet operator

The price control is another method that can be adopted by the
FO, as it is shown in Fig. 6. This control method requires forecast of
EV users' response to the prices sent by the FO. The price signal
can be designed as time-of-use price or dynamic prices.
Depending on the aggregation objectives presented in Section

2.2, e.g., different objectives have different requirement on EVs

in term of response time;
Depending on the EV consumer's participation. Some consumers do not like their EVs to be controlled by FOs. Under such
circumstance, transactive control and price control are suitable
methods;
Depending on the business model, e.g., whether the economic
benets of optimal charging of EVs can justify the cost of
communication infrastructure in all control cases.
4.4. Discussion on integrating the control strategies
Although most research assume either centralized control or

decentralized control including transactive control and price
control architecture, this is indeed an important decision which
should be taken in the earlier stage. From our perspective, three
issues shall be investigated:
The authors in [76,77] compared the centralized control and

decentralized control method when using them to make an plan
for optimal delivery of energy to EVs as well as avoiding grid
1216
agent systems technology, it is possible to model the optimization

and the control occurring in the smart charging of EVs.
5. Mathematical modeling and control algorithms: centralized

control of eet operator
In this section, the algorithms often used in the centralized
control are presented. Linear programming, quadratic programming, dynamic programming, mixed-integer linear and non-linear
programming, stochastic programming, robust optimization,
heuristic optimization and model predictive control will be discussed through an extensive literature review. Further, a qualitative comparison among the nine types of algorithms will be presented at the end of this section.
5.1. Linear programming
Fig. 5. Information ow between the FO and the EVs in a transactive control

strategy.
Sundstrom and Binding [78], used linear approximation to

characterize the state of charge of a battery and formulated the
charging process of an EV eet into a linear programming (LP)
based optimization problem:
min t s cT P b
Subject to
8
As P b Z bs
>
>
>
>
< Ag P b Z bg
Ab P b Z bb
>
>
>
>
: b r P r bu
l
b
With the time slot t s , price vector c, the charging power P b

(decision variable), the stopover inequality constraints (As, bs), the
generation inequality constraints (Ag, bg), the battery inequality
constraints (Ab, bb), and the upper and lower power bounds (bu, bl).
The solution of this linear optimization problem is the optimal
charging prole that minimize the charging cost of EV eet [78].
5.2. Quadratic programming
Fig. 6. Information ow between the FO and the EVs in a price based control
strategy.
congestions. They outlined the advantages and disadvantages of

both strategies, mainly from the communication perspective.
However, more research is needed to evaluate how the choice of
control strategy inuences the overall performance and engineering requirements (e.g. information, communication and
computation requirements).
When implementing different control strategies of smart
charging of EVs, especially transactive control method, it is
recommended that multi-agents system based technology is very
suitable to design a coordinated and collaborative system for a
smart charging network of EVs. In multi-agent systems, different
interests of various actors shown in Fig. 1 can be presented and
coordinated by using smart charging method. By using multi-
A nonlinear approximation (quadratic formulation (QP)) of the

battery charging model is studied in [78]. The results showed that
the number of constraints and calculation time are higher and
they increases faster with a growing eet in the quadratic formulation than in the linear formulation. An example is conducted
for comparison and the result indicated that calculating time using
the quadratic formulation is 819 times the calculation time using
the linear formulation [78]. But the result difference does not
justify the benets of using quadratic formulation. Recently, linear
and non-linear programming algorithms applied to renewable
energy are summarized and compared in [79]. In [80], ClementNyns et al. formulated the power losses problem caused by large
penetration of EVs in the grid into a sequential quadratic optimization problem. The charging power obtained by the quadratic
programming cannot be larger than the maximum power of the
charger P max . The batteries must be fully charged at the end of
cycle, so the energy which ows to the batteries must be equal to
the capacity of the batteries C max . xn is zero if there is no EV
connected and is one if there is an EV connected at node n. The
above problem specication can be represented as follows:
min
t max X
lines
X
t 1l1
Rl :I 2l;t
Subject to

8
8 t; 8 n A nodes :0 r P n;t r P max
>
>
>
>
t max
<

X
8 n A nodes :
P n;t :t:xn C max
>
>
t1
>
>
:
xn A 0; 1
where Rl represents the resistance of line l, I is the current in line l

in period t, P n;t is the active power of EVs charge in the bus n in
period t, P max is the maximum charge rate and C max represents the
battery capacity. Finally the charging control is imposed by the
binary variable xn .
The quadratic programming techniques are applied using both
deterministic and stochastic methods [80]. The input variables in
both cases are the daily/hourly load prole. In the deterministic
case, the load proles are static. In the stochastic case, the load
prole are transformed into probability density functions, which
means the xed input parameters are converted into random
input variables with normal distributions assumed at each node.
The details of stochastic case are presented in Section 5.6.
5.3. Dynamic programming
Dynamic programming (DP) is widely used for different purposes in electric vehicle smart charging problem. In this study, the
work in [31] is introduced where the purpose is to minimize
charging cost by participating in regulation market. Firstly, a
specic control strategy is denoted by
u0 ; u1 ; uk ; ; uN 1
where uk is the control variable for time k that denotes a dimensionless and discrete representation of P k . P k corresponds to the
purchased power. The total cost of a whole charging sequence, J is
calculated as:
J 0 x0 J N xN
N
1
X
lk xk ; uk ; k
k1
J N means cost of the nal step, lk xk ; uk ; k denotes the cost-to-go

for all other steps, N denotes the total number of time intervals, xk
is the state variable of the battery. The objective is to nd the
optimal control variables that can minimize the total cost. The
detailed mathematic formulation of cost of nal step and cost-togo are not presented here. The purpose of the function used for
calculation of cost of nal step is to ensure that the battery is fully
recharged before the rst trip of the following morning. For the
function of cost-to-go, the electricity price, regulation-up price
and regulation-down price are considered.
This is a classical dynamic programming formulation and the
optimal trajectory is calculated starting with the cost of the last
state and going backwards through time until the rst states
optimal cost J 0 x0 is given by the algorithm. Concerning the
computing time of dynamic programming, the results in [80] show
that the difference of the charging proles for the QP and DP
technique are negligible, however, due to heavier storage
requirements of the DP technique compared to the QP technique,
hence, the computational time for DP technique is longer.
5.4. Mixed-integer linear programming
The use of mixed-integer linear programming (MILP) techniques is necessary when binary/integer variables are introduced in
the problem. Several components in power systems have discrete
characteristics, such as capacitor banks, transformer tap changers
and thermal generators [81]. In electric vehicles charging problem,
the binary variables can be used to determine the EV state namely:
charge, discharge, and driving [82]. In [83], MILP is used to
1217
determine the impact of EVs in a power systems with high

penetration of wind generation and demand response programs.
In [84], MILP is proposed to determine the optimal EVs charging in
unbalance distribution networks.
The most used optimization techniques to solve MILP are
branch-and-bound, cutting-plane and branch-and-cut methods
[85]. Branch-and-bound is the most widely used method for solving MILP [85]. The method consist of an implicit enumeration of
candidate solutions to the MILP problem (forming a tree), and then
each branch of the tree is explored and checked against a bound of
the optimal solution. If the branch cannot nd a better solution
than the best one found so far (or bound), the branch will be no
more explored. Each candidate solution is obtained by solving a
relaxed sub-problem of the original MILP problem (by relaxing the
integer conditions), and each branch contains additional constraints that limits the range of the integer variables. These additional constraints will help to obtain new candidate solutions with
integer values in these integer variables. On the other hand, the
cutting-plane method iteratively solves the non-integer version of
the MILP problem by including linear inequalities (or termed as
cuts) to dictate that the integer variables assume integer values
[85]. Currently, the branch-and-cut is the best choice to solve the
MILP, it follows the same procedure as branch-and-bound and
meanwhile incorporating the cutting-plane method [85]. This
incorporation helps the branch-and-cut technique improving the
computational performance (CPU time and memory) when compared with the branch-and-bound. Additionally, the branch-andcut can incorporate relaxation and decomposition methods (e.g.
Lagrangian Relaxation and Bender's Decomposition) to deal with
large MILP problems.
To demonstrate the formulation of MILP problem, the example
of Section 5.2 is used, considering the EV maximum and minimum
charging/discharging limits in each bus. Besides, EVs can only
charge or discharge in each time period at same bus. Additionally,
the objective should be a linear one, such as the energy cost (1)
presented in Section 5.1. In MILP formulation, the constraints (4)
presented in Section 5.2 should be adapted, in which the binary
variables are introduced as following:
8

8 t; 8 n A nodes :xc;t :P min rP c;t rxc;t :P max
>
>
>

>
>
8 t; 8 n A nodes :xd;t :P min r P d;t r xd;t :P max
>
>
>

>
>
>
8 t; 8 n A nodes :xc;t xd;t r1
>
<
7
>
t max
>

X
>
>
>
P c;t P d;t :t C max
>
> 8 n A nodes :
>
>
t1
>
>
>
:
x A 0; 1
n
In that case, the Pmin has been dened as zero for the charging
and discharging process, but in this new formulation the charging
power Pc,t and the discharging power Pd,t can be 0 or assume a
value between the minimum and maximum. The binary variables
xc,t and xd,t have been introduced to determine when (time period)
the EVs are charging or discharging.
5.5. Mixed-integer non-linear programming
The mixed-integer non-linear programming (MINLP) is used in
the EVs scheduling mainly when other distributed resources are
considered due to their non-linear operation characteristics, and
when it is necessary to introduce the network technical constraints such as the lines thermal limits and the bus voltage
operation boundaries [86]. The objective functions can be technical or economic taking into account the power losses [87], the
operation costs [82], the load balancing [37], the voltages proles
[88], the greenhouse gas emissions [48] among others.
1218
The introduction of the network constraints require the

determination of the power ows in the network through its DC or
AC models. The AC model is the main challenge due to its complex
formulation of Kirchhoff's law for determining the active and
reactive balance in each bus, that can be formulated as:
P Git P Dit Gii V 2it V it
X

V jt Gij cos ijt Bij sin ijt
j A Li
Q Git Q Dit V it
X
jAL

V jt Gij sin ijt Bij cos ijt Bii
V 2it 8 t A f1; ; T g;i A f1; ; N B g;ijt jt jt
where V it and it are the voltage magnitude and angle in each bus i
and in each period t, respectively. The lines characteristics are
represented by the variables Bij , and Gij (imaginary and real parts of
the admittance matrix corresponding to the i row and j column).
Finally, generated power from different resources are represented by
P Git and Q Git and the consumption power including loads and
storage systems charge (EVs and storage systems) are represented by
the P Dit and Q Dit . Additionally, the MINLP formulation includes the
same integer variables that are used in MILP.
In general, the use of MINLP is suitable for problems with small
number of EVs and other resources (small number of variables).
For a large number of resources, the computational requirements
and the execution time turn the use of this methodology inadequate for real application. But, the results obtained are more exact
and can takes into account more realistic models of the devices
connected to the network. This means that the MINLP can be used
in small cases to determine the errors of each approximation made
by other techniques [82] or in real applications when a problem is
restrict to a small area with few number of resources.

E v^ n can be estimated by generating M independent samples
i;j of the random input variable each of size N. The M optimization
problems are performed where the nonlinear power ow equations are solved by using the backward-forward sweep method
[91]. The optimal values of M samples constitute a normal distribution:
(
)
N

1X
i;j
^v jn min f^ N P n;t
gP n;t ; ; j 1; ; M:
12
Ni1
j
v^ n is the mean optimal value of the problem for each of the M
samples. LN;M is an unbiased estimator of Ev^ n . Simulations indicate that in this type of problem, the lower bound converges to the
real optimal value when N is sufciently high:
LN;M
M
1 X
j
v^
Mj1 N
13
A forecasted daily load le for the next 24 h is required in the

beginning, then the daily prole of the available set are varied by a
normal distribution function. The standard deviation is determined in such a way that 99.7% of the samples change at maximum 5% or 25% of the average. In general, the simulation results
indicated that the difference between the power losses of the
stochastic and the deterministic optimum is rather small.
Other studies such as Fluhr et al. [92] uses Monte-Carlo method
to generate the probability distributions of the travel paths for one
week with the survey Mobility in Germany (MIG), because the
original data MIG only provide one day driving behavior; study in
[93] uses normal distribution and Poisson distribution to investigate the probabilistic distribution of plug-in time and initial state
of charge of EVs. A fuzzy set model is used in [89] to deal with
uncertainties related to electricity market prices and ancillary
service deployment signals.
5.6. Stochastic programming

5.7. Robust optimization
Most of the current researches [31,78] assume that the load
proles, initial state of charge, driving pattern, grid conditions and
electricity price are known and deterministic. However, this is
certainly not the case in a realistic scenario. Therefore, there is a
need to adopt stochastic approach to reduce the risks introduced
by the uncertainties related to mentioned aspects. Recently, some
articles [89,80,48] have been published with stochastic methodologies to deal with uncertainties in the EVs management.
A stochastic approach is considered in [80] when minimizing
the power losses problem. A sample average approximation
method [90] is applied to formulate the random inputs and the
lower bound estimate principle is used to estimate the optimal
value. It is noted that the model is the same as presented in Eq. (3)
of this section (Section 5.2). The uncertainties of these parameters
are formulated as probability density functions, in which the xed
input parameters are converted into random input variables with
normal distributions assumed at each node. N independent samples of the random input variable j , the daily load prole, are
selected. Eq. (10) gives the estimation for the stochastic optimum
v^ n . The function gP n;t ; j gives the power losses and P n;t is the
power rate of the charger for all the EVs and time steps. f^ N is a
sample-average approximation of the objective of the stochastic
programming problem:
8
9
N
<

=
1X
j
^
v^ n min f N P n;t
g P n;t ;
10
:
;
Nj1
The mean value of the power losses Ev^ n , is a lower bound for
the real optimal value of the stochastic programming problem, v ,
Ev^ n rv
11
Robust optimization is other suitable method to handle the

variables uncertainties [94]. The rst step leading to robust optimization was given in 1950s. The main goal was to construct
worst-case distributions for well-structured problem classes. The
robust optimization came from the robust control community
being initially proposed by [95]. But only in the last nineties it has
been used in real problems and applications. The robust optimization considers that the uncertainties model is deterministic and
set-based. One of the most important characteristics of robust
optimization is that the obtained solution should be feasible for
any realization of the uncertainty in a given set [94].
The robust optimization method can be used in power systems
in several optimization problems due to the consideration of
uncertainties in consumption, generation, market prices, and
electric vehicles requirements [96]. For example, the adaptive
robust optimization is used in [96] to solve the economic dispatch
and in [97] to solve the security constrained unit commitment
problem. In [98,99], a robust optimization model is used to solve
EV charge/discharge scheduling problem. In [98], both the electricity network and the transport sector constraints are considered
in the model which is applied to the real case of Ontario, Canada,
to determine Ontario grid potential for supporting EVs for the
planning horizon 20082025. In [99], the robust model also
includes the distributed generation constraints.
Taking into account the use of robust optimization in a generic
economic dispatch presented in [96], it is possible to divide the
problem into two stages represented in the objective function

minp Z 0;R;pd cTp P cTR R maxdD cT pd
14
In the rst stage, the dispatch cost and the regulation capacity
cost are considered. In the second stage, the worst case
performance-based regulation cost is used. The variables P and R
represent the energy dispatch decision and the regulation amount
for automatic generation control (AGC) units presented in the rst
stage, respectively. The variable p(d) represents the secondstage energy output from the regulation unit, which is adjustable
to each demand realization d d, for d A D, in which D represents uncertainty set. In the economic dispatch problem, the
electric vehicles can be considered as a load in charge mode and as
generator in discharge mode in the objective functions. The
technical aspects of the EVs are considered in the constraints that
subject to the objective functions.
5.8. Heuristic and meta-heuristic algorithms
One of the main problems of the EVs scheduling is its interdependency between periods. For instance, the energy stored in
the battery in a certain period depends on the decisions applied in
previous periods, such as charging/discharging power or energy
spent in trips, and also on the decision applied in that period.
Besides, a large-scale penetration of EVs can increase signicantly
the number of variables and consequently the execution time to
nd an optimal solution in an optimization problem. Considering
the complexity, e.g., in [86], a deterministic technique (mixedinteger non-linear programming) is implemented to solve the dayahead resource scheduling considering a penetration of 2000 EVs
and this technique took around 28 h to obtain the optimal solution. In this kind of problems, it is easily that a large number of
variables and constraints is reached, such as 100,000 problem
variables in [100]. This problem can be classied as a NP-hard
problem [101]. The NP-hard problems include, no polynomial time
algorithms, i.e. algorithms that need exponential computation
time in the worst case to obtain the optimum solution [102].
The heuristics optimization algorithms intend to solve these
problems obtaining a solution near to the optimal (not necessary
the optimum) one in a convenient execution time. For example, in
the previously mentioned study [86], the proposed meta-heuristic
took around 18 min to nd a solution with a difference close to the
optimal one in 0.97%. In order to adapt the heuristics to real problems and to improve their performance, it is necessary to include
some additional algorithms in some steps of the heuristics process
that results in meta-heuristic algorithms.
An extended overview of the population-based meta-heuristic
techniques is presented in [103]. Regarding the use of metaheuristic in power systems, a review is presented in [104].
Recently, meta-heuristic methods applied to renewable energy is
reviewed in [102]. Thus, the following paragraphs intend to
overview the application of some meta-heuristic techniques in the
EVs charge and discharge scheduling.
In [100], a variant of PSO is proposed for solving the EVs charge
and discharge process that minimizes the operation cost of electric
vehicles. The main feature of the proposed method is the inclusion
of a process to change the particles velocity in each iteration.
Articial bee colony (ABC) is used in [105] to optimize the management of distributed energy resources, including distributed
generation, demand response and EVs. An ABC-based multiobjective algorithm is proposed in [106] to optimally determine
the number, locations and sizes of the distributed generation and
parking lots.
Genetic Algorithms (GA) is used in [107] to optimize the EVs
charge scheduling that aims at minimize the operation costs while
taking into account the network supply constraints. In [108], a
modied GA algorithm is used to optimize the PHEVs charge
patterns that consider multi objectives: 1) the total cost of fuel and
electricity and 2) the total battery health degradation over a 24 h
1219
period. The papers concluded that these two objectives are conicting which resulting in a Pareto front optimal charge pattern.
Simulated annealing (SA) is proposed in [109] to schedule the
EVs charging and in [110] to schedule the charge and discharge
process, including the AC power ow constraints. One of the main
disadvantages of SA is its dependency of the initial solution.
Hence, in order to overcome this characteristic, a hybrid approach
is proposed in [86] using an ant-colony optimization (ACO) algorithm to determine the initial solution that will be used by SA. The
same authors [111] proposed two heuristics to obtain a good initial
solution for the EVs and distributed generation, respectively. These
heuristics presented an initial solution worse than the ACO in [86],
however shown a much lower execution time than the ACO
algorithm. Beyond the mentioned use of ACO to provide the initial
solution to SA proposed in [86], a hybrid PSO and ACO algorithm is
proposed in [112] to deal with short-term unit commitment problem with EVs considering different reliability limits.
5.9. Model predictive control

Model predictive control (MPC) is a control algorithm that
optimizes a sequence of manipulated variable adjustments over a
prediction horizon. The main advantage of MPC is the fact that it
allows the current timeslot to be optimized; meanwhile it takes
future timeslots in account. This is achieved by a nite timehorizon optimization, but only implementing the current timeslot.
MPC has the ability to anticipate future events and can take control actions accordingly. The use of MPC in EVs smart charging is
proposed in [113], considering the cost minimization objective
function. The model takes into account the EV drivers' preferences,
technical bounds on the control action (the charging rate is
modeled as a semi-continuous variable) and both market and grid
constraints. In [114], a MPC based power dispatch approach is
proposed to minimize the operational cost while accommodating
the PEV charging uncertainty. In [115] the economic MPC is used
for electric vehicle charging planning. The model considered that
the EVs can be used for both peak reduction and for ancillary
services, by absorbing short term variations in the electricity
production. According to the study, the use of proposed model can
lead to 5060% of savings when compared to uncontrolled charging. A two-stage hierarchical MPC model is proposed in [116]. On
the upper level, the model considers the voltages deviations
through the control of the generators. On the lower level, the
model controls the EVs charging taking into account the driving
requirements. The MPC method is also used in some experiments
[117]. The experiences are made under the context of the
SMARTV2G project and the focus is the implementation of a
centralized demand side management using MPC control algorithm, which allows remote real time control of the charging
stations considering the players preferences.
5.10. A summary of the presented algorithms with three types of

criteria
In Table 5, the information of the presented algorithms in terms
of computation time, certainty of performance and applicability is
summarized. The summary aggregates the comparisons described
in the literatures in term of execution time and performance of the
presented algorithms. The applicability of the presented algorithms is summarized from two perspectives.
1220
Table 5
General comparison between the presented algorithms.
Execution time
Certainty of performance
Applicability in general
Applications to EV charging
Linear programming used in:

[78].
Generally, it is the fastest one.
Results in [78] showed that the performance is

excellent in term of nding the optimal solution
Minimize charging cost of EVs.
Quadratic programming used

in: [78,80].
Ref. [78] showed that the calculation time using the Ref. [78] showed that the difference between using
QP is 819 times than the one using LP for a eet of LP and QP is minor. Therefore, the benet of using
the QP does not justify the increase in computation
50 vehicles.
time.
Ref. [80] indicated that the computational time for Ref. [80] showed that the difference between the
DP is slower compared to QP.
charging prole of using QP and DP is negligible,
although the QP gave more accurate results.
1) The objective function is linear, and the set of

constraints is specied using only linear equalities
and inequalities.
2) Standard model, easy for implementation.
1) The objective function has quadratic terms, while
the feasible set must be specied with linear equalities and inequalities.
2) Standard model, easy for implementation.
1) Studies the case in which the optimization strategy is based on splitting the problem (EV charging
schedule) into smaller sub-problems (multi-time
slots).
2) No standard model, difculty increases for
complex problem.
3) Give global optimal result.
1) Studies considering a short time to obtain the
solution.
2) Model the discrete operation of some power
components (e.g. capacitor banks).
3) Start-up and shut-down costs.
Dynamic programming used

in: [31,80].
1) Minimize charging cost of

EVs.
2) Minimize power losses of
power systems.
EVs.
power systems.
3) Maximize prot of providing regulation services.

EVs.
2) Minimize losses of power
systems.
1) Studies considering the network technical
Ref. [82] shows that the operation cost increase
Mixed-integer non-linear pro- Ref. [82] indicated that the computational time is
EVs.
due to the losses are considered in the model. The constraints.
gramming used in: [8688]. highly dependent of the number of EVs. Other
2) Tested in single and multi-objective functions.
important aspect is if the problem considers the DC main advantage is the network technical conpower systems.
straints validation.
or the AC network model.
3) Give global optimal result.
4) Load levelling.
Studies the case in which some of the constraints or 1) Minimize charging cost of
Stochastic programming used
The computation time is longer generally because
The simulation results in [80] indicated that the
EVs.
parameters (Load prole, driving pattern etc.)
in: [80,92,93].
more scenarios are considered.
difference between the power losses of the sto2) Minimize power losses of
depend on random variables.
chastic and the deterministic optimum is rather
power systems.
small.
Studies the case in which some of the constraints or 1) Minimize charging cost of
Robust optimization used in:
As in stochastic optimization, the computation time According [96], the robust optimization is less
EVs.
parameters (Load prole, driving pattern etc.)
complex to implement in real problems than the
[9699]
is longer generally because more scenarios are
depend on random variables.
stochastic optimization. However, the stochastic
considered.
power systems.
optimization is more mature technique, tested to
3) Maximize prot of providsolve several problems.
ing regulation services.
In [86] and [100], there is solved optimization
The exaction time is lower when compared with
Heuristic and meta-heuristic
1) Studies considering the network technical
problems with different scenarios and the metaother techniques. However depends a lot of the
optimization used in:
constraints.
EVs.
used technique and also the required results quality. heuristic results presented a small error lower than 2) Tested in single and multi-objective functions.
[86,100112].
3) Tested in problems considering the
power systems.
1% in comparison with the optimal solution.
uncertainties.
3) Maximize prot of provid3) Use of several techniques and several hybrid
ing regulation services.
methods.
4) Load levelling.
4) Give a local optimal result.
Method used for studies in scheduling and near-real 1) Minimize charging cost of
In [113], the MPC method obtained excellent
Model predictive control used This technique presents a low execution time,
EVs.2) Minimizing the netresults in a rolling scheduling problem, considering time operation, considering some uncertain
in: [113117].
therefore it is a suitable approach to be used as
work voltage deviations.
parameters.
different predictions in the method, such as EV
control method in near-real time operation.
drivers preferences.
Mixed-integer linear programming used in: [81,83,85].
The execution time depends substantially on the

number of binary and integer variables.
The objective function is similar to the LP and QP

problems when the same constraints are considered. However, the binary and integer variables
can change signicantly the operational limits.
Control algorithms
6. Mathematical modeling and control algorithms: transactive

control of eet operator
The following papers [52,71,118,119122], are selected to
illustrate the transactive control's application in EV eet management. In [52], transactive control method is applied to solve
distribution network grid congestion between the distribution
system operator and the electric vehicles eet operators. Firstly, EV
eet operators formulate a cost function that reects the charging
power deviation from the scheduled charging power, e.g., in the
form of a quadratic function:

2
15
k;i C k;i P~ k;i P Ek;i
where k is the index of EV eet operator, i is the index of time slots.
P~ k;i is the decision variable, P Ek;i is the scheduled charging power
and it is the optimal charging schedule for EV FO. However, the
sum of the charging schedule P Ek;i may bring operational challenge
to DSO and thus it needs to be modied. The overall objective is to
minimize the cost function of EV FOs with respect to the grid
capacity constraints. The minimization is formulated as a Lagrange
problem and solved iteratively using a decomposition algorithm.
The Lagrange multipliers are interpreted as congestion price that
coordinates the EV FOs charging proles. Furthermore, the study
is extended in [118] to solve the voltage band violations by
introducing congestion prices on the buses level.
In [71], Ipakchi pointed that a higher penetration of distributed
resources will require a greater attention to distribution congestion issues and a need for improved distribution automation and
distribution management capabilities. A transactive control
approach is proposed to solve the problems. In the example
described in [71], a plug-in electric vehicle requests using 7.8 kW h
of charging energy over the next two hours. This request can be
presented as a demand transaction and sent to a demand-side
management application operated by the utility distribution
company. Knowing the transaction delivery point to which the car
charger is connected to, this application will check the available
capacity of the secondary distribution transformer, lateral and
feeder circuits. Then it determines whether the additional load
will impact the circuit reliability and cause any adverse phase
imbalances. The demand-side management application will then
schedule the charging for the requested time period. At the same
time, the application may receive many more information such
charging requests that have to be checked, and in aggregate they
have to be coordinated with wholesale scheduling at substation
supplying the feeders to ensure adequate supply. Each of these
actions could be modeled as a transaction between a consumer
system, a utility decision support system, distribution eld
equipment and supply scheduling system in an aggregate form.
Ma et al. [119] formulate a class of nite-horizon dynamic game
(or a transactive control system) to optimally control the charging
prole of a large-scale of electric vehicles. Within the game, the
control objective is to minimize electricity generation costs by
establishing an EV charging schedule that lls the overnight
demand valley. Moreover, the paper establishes a sufcient condition under which the system converges to the unique Nash
equilibrium. The key formulas are listed below:
xnt 1
n
xnt n unt ; t
0; ; T 1
16
where xn is the state of charge of EV n , n and n means the charging efciency and battery size of EV n , and un represents the local
control variable. The purpose of the study is to nd the set of
feasible full charging controls, which are described as follows:

17
n un un0 ; ; unT 1 ;s:t:unt Z 0; xnT 1
1221
where the nal constraint on xnT requires that all EVs are fully
charged by the end of the interval. The cost function of agent n,
denoted by J n u is used as criteria and specied as:
J n u
TX
1n

2 o
pr t unt unt avg ut
18
t0
where each agent's optimal charging strategy must achieve a

trade-off between the total electricity cost pr un and the cost
incurred in deviating from the average behavior of the EV population un avg u2 . With these criteria and certain conditions, the
theorem about the existence of the Nash equilibrium is presented
in the paper [119]. The proposed algorithm ensures convergence to
a at, or optimally valley lling aggregate charging prole. In order
to implement the transactive control system, an iterative algorithm for computing the unique Nash equilibrium is proposed
which includes four steps: 1) the utility/eet operator broadcast
the forecast of base demand to all the EVs. 2) Each of the EVs
proposes an optimal charging strategy that minimizes its charging
cost with respect to a common aggregate EV demand broadcasted
by the eet operator. 3) The FO collects all the optimal charging
strategies proposed by the individual EVs, and updates the
aggregated EV demand to all EVs. 4) Repeat step 2) and 3) until the
optimal strategies proposed by all EVs no longer change. Similar
study is performed in [120] where Gan et al. further proved that
transactive control based algorithm converges to optimal charging
proles, irrespective of the specications of EVs, even with asynchronous computation. Besides, the authors also extended the
algorithm to track a given load prole and to real-time
implementation.
In [121], a scalable three-step approach for demand side
management of EVs is presented. The three steps consist of
aggregation, optimization and control. In the aggregation step,
individual EV charging constraints are aggregated upwards. In the
optimization step, the aggregated constraints are used for scalable
computation of a collective charging plan, which minimizes costs
for electricity supply. In the real-time control step, the calculated
charging plan is used to create an incentive signal for all EVs,
determined by transactive control method. These three steps are
executed iteratively to cope with uncertainty and dynamism. The
modeling method of the real-time control step is presented as
follows:
1) The individual power constraints of an EV i at time t0 are
represented in a demand vector iPdem, which contains all possible power values for charging the EV's battery. These power
values vary between iPmin and iPmax, and are specied by a selfdened piecewise linear function ifd.
8
i

i
< P i r P max r P min ; 0 r r o i r
i
c
max
i
c
fd
19
: i
i r
P min ;
min r r o rmax
i dem
i dem i dem
P r j P r

o
f d r ; 8 r A 0; ; rmax
20
where iprc the corner price of EV i, and the corner price of EV i is

a heuristic which determines the slope of a demand function.
2) Based on the piecewise linear function ifd dened in (19), the FO
uses the aggregate demand functions and the optimal charging
decision obtained in optimization step, i.e., P 0 in the original
paper to generate the equilibrium price req .
3) This equilibrium price is sent to all EV agents. Upon receiving the
signal, each EV agent will locally match req in its own demand
vector, which amounts to the individual charging power.
Distributed utility maximization

problem.
[122]
[121]
[120]
Non-cooperative game theory

method.
Optimal control theory with convex
function.
Piecewise linear cost function.
[119]
A decentralized computational algorithm is developed to

reach the convergence.
A decentralized algorithm is proposed to iteratively solve the Aims for at loading prole.
optimal control problem.
Dynamic programming.
1) Minimize charging cost.
2) Prevent overloading problem.
A general multi-agent framework is used.
Primary and Secondary Frequency
control.
Auction based decision making.

[71]
The existence, uniqueness and optimality of the Nash equilibrium of the

EV charging problems are studied.
Compared with [119], it is proved that asynchronous computation can
also converges to optimal charging prole.
Iteration of the three-step approach is proposed to address the uncertainty of EVs charging plan.
Devices can identify their value of the received energy using utility
functions.
1) Prevent distribution grid

congestion.
2) Solve Voltage problem.
3) Minimize charging cost.
Prevent distribution grid
congestion.
Minimize electricity cost.
Dual decomposition algorithms.
A hierarchical control structure is proposed that includes DSO, FO, and
EVs.
Convex optimization modeling.
[52,118]
Merit-order based method.
Control/computing algorithms
Key features
References Modeling methods
Market platform is needed at the distribution system level.
Table 6
General comparisons among the presented studies.
1222
In [122], the auction based transactive control is applied to

control the cluster of loads with the purpose of providing spinning
reserves. Firstly, each device denes an utility function for the
utilization of the electricity, e.g., the corner price model developed
in [121] is applied for calculating a bid function of an EV. Then, in
real-time operation, all the device agents send their bid to a
concentrator agent or eet operator agent. The concentrator agent
sums up the bid functions of their zone and then sends the
aggregate bid function to a unique auctioneer agent. Finally, the
auctioneer agent will dene the equilibrium price as the intersection of the aggregate bid functions and the supply bid function.
After the equilibrium price is dened, it is sent back to all of the
devices agents and the corresponding power of the device agents
will be determined. The market clearing takes place every 15 min
or can be made event-driven. Furthermore, the transactive control
method is extended to cooperate in frequency reserve markets.
In summary, a key operational parameter used in transactive
control is value (i.e., cost/utility functions in [52,71,118,119122])
and thereafter the equilibrium price can be discovered and the
transaction can be executed. It is seen that iterative information
exchange is required to reach equilibriums between the eet
operator and electric vehicles in [52,118120], while only one-time
information exchange is required in [71,121,122] to reach the
equilibrium. Furthermore, the modeling methods and algorithms
to implement the transactive control for electric vehicles integration are summarized and compared in Table 6.
7. Mathematical modeling and control algorithms: price control of eet operator

Price signal used in the price control ranges from time-of-use
electricity rate/tariff [74,123] to hourly varying prices [75,124].
Shao et al. [74] focused on the development of demand response
model for residential customers with EV penetration that reects
customer behaviors in response to variable electricity prices. Nine
types of residential customer loads are divided into three groups:
critical, interruptible and deferrable loads. Of which, the deferrable
load means it shuts down the equipment when price is higher
than a pre-determined value, the load will be shifted to lessexpensive hours. EV ts into the deferrable load type. It is
assumed that if the time-of-use price is increased by 100% from its
corresponding at rate, 20% customers are willing to shift or shed
their non-critical load. For simplicity, the participation function is
linear. Two EV penetration levels are studied consist of low
penetration level and high penetration level. There is one EV for
every ve houses in the low penetration level and there are two
EVs for every ve houses in the high penetration scenario. With
the time-of-use electricity rate, it is shown in the paper that the
time-of-use rate helps reducing the peak load. Note that if the EV
users use timing device to largely avoid the peak tariff and charge
it in less-expensive hours, rebound effect [125,126] may exists due
to the time-of-use electricity tariff. As discussed in [127], since
there is no inertia in EV charging process, the rebound comes
faster and the peak value depends on the previously curtailment
period and the size of curtailed energy. The problem can be solved
by a well-designed time-of-use electricity tariff and diversied EV
responsive strategies.
The relationship between EV charging behavior and time-ofuse rates are specically explored in [123] where suggestions are
made for conducting a well-designed pricing experiment. The
purpose of the experiment is to determine whether such rates
help reducing future grid reliability problems as EV penetrate in
the vehicle market. In [123], the EV user will charge the vehicle 25
times during the month based on assumptions of a monthly travel
miles of 1250 and each charge lasts four hours. Three types of
1223
Table 7
General comparisons among the presented studies.
References Modeling methods
Key features
[74]
Household loads and EVs charging prole are studied.
[123]
[75]
[124]
Three types of load shedding and

shifting strategies are dened logically.
Conditional logit model.
Price signal
format
Time-of-use
tariff.
A social experiment is suggested to estimate the price response of Time-of-use
EVs' charging.
tariff.
1) Parametric utility model.
Compared to [74], utility functions are developed for the three
Dynamic price
2) Stochastic expectation model.
types loads: critical, interruptible and deferrable load.
signal.
Dynamic price
Distribution locational marginal pricing It is discussed that compared with linear programming based
signal.
method via QP.
DLMP formulation [128], QP based formulation ensures convergence to the DLMP.
time-of-use rates are used and each rate has three pricing periods:
peak, mid-peak, and off-peak. In the simulation study, the authors
developed a conditional logit model to get the choice probabilities
of EV charging in presences of different time-of-use rates. The
simulation results shown that a value of 0.80 will be needed to
effectively eliminate peak time charging introduced by electric
vehicles' charging and a value of 0.25 will be needed to eliminate half of the normal peak time charging load, after running
simulations with a wide range of price elasticity.
Yu et al. [75] investigated the price elasticity of electricity
consumers and these are also the important aspects in price
control method. In [75], the marginal utility function of loads is
realized by the following parametric stochastic process
8
< t ; r t r ;
rt
21
otherwise:
: 0;
where ; ; ; are random variables that describes the different
characteristics of utility function, stands for the time slot that a
task is initially requested, which also reects the task distribution,
is the initial marginal utility, which stands for the magnitude of
the marginal utility, is the tolerable delay, which determines the
maximum delay that a user can tolerate to nish a task, means
the utility decay rate, which represents the cost of inconvenience
by the delay.
Using the model, the scheduling of each individual task is now a
random event whose probability distribution is controlled by the
stochastic process rt. The aggregate demand curve can be estimated through expectation with respect to the distribution of rt.
Note that some assumptions have been made in [75], such as the
time period of the scheduling is divided into T time slots, there are
total M individual tasks m: m1,, M of different appliances that
are to be initialized by all the users within the scheduling period,
and each task will consume xm kW h. Furthermore, it is assumed
that each task can be completed within one time slot; therefore,
tasks that have duration longer than one time slot will be decomposed into multiple tasks that are considered independently.
Huang et al. [124] proposed a distribution locational marginal
pricing (DLMP) method through quadratic programming designed
to alleviate congestions that might occur in a distribution network
with high penetration of exible demands such as electric vehicles
and heat pumps. In the DLMP method, the DSO calculates dynamic
tariffs and sends them to the eet operators. The eet operators
make the optimal energy schedules for the exible demands. It is
proven in the study that the DLMP through quadratic programming modeling can ensure a unique marginal price for the FOs.
Instead, DLMP through linear programing modeling [128] might
bring multiple solutions issue of the eet operator optimization
that may cause the congestion management by DLMP to fail. The
case studies showed that DLMP helps alleviating congestion problem. It is also shown that the linear programming based modeling can lead the failure of the DLMP.
Reduce the peak demand.

Avoid transformer overload.
Reduce the transformer
overload.
Manage distribution network congestions.
In summary, studies [74,75,123,124] suggested that the electricity price can be properly designed to reduce the peak demand
as EVs penetrate in the vehicle market. However, it is also noted in
[74,75,123] that to which extent the properly designed price signal
could assist in maintaining grid reliability will remain open until
empirically tested EV owner's price responsiveness through
experiment pilots are known. When investigating EV owner's
price responsive behavior, decomposition methods [26] widely
used in energy economics eld can be applied to nd the relation
between EV's charging response and electricity price, state of
charge, customer driving habits and reasons for buying the EV etc.
Additionally, Table 7 depicts the information of the presented
algorithms in terms of modeling methods, key features, price
signal format and applications to EV charging. The presented
algorithms are also summarized and compared in this table.
8. Conclusion and recommendations

8.1. Conclusion
As a conclusion, it is learned from this study that centralized
control, transactive control and price control show their pros and
cons for management of an electric vehicle eet. Centralized
control offers best performance in controlling the EV eet and
therefore making better optimized charging proles. For the centralized control, a linear programming based technique is recommended to characterize the optimal charging problem and generate the optimal charging schedule. In order to generate an
optimal charging schedule, a forecasted electricity price and predicted EVs driving pattern are essential. Fortunately, they can be
estimated by the commercial actors. Nevertheless, EV owners are
encouraged to submit a provisional EV utilization plan for next day
to the commercial actor for generating an optimal charging
schedule. For the centralized control to become a success, more
research is needed in setting up a collaborative business model
which ensures the proper engagement of commercial actors and
EV owners. Transactive control shows its advantages such as
providing controllability of the EVs to the end-users and scalability
of implementing the method in a large scale of electric vehicles.
However, for the transactive control to become a success, some
barriers needed to be resolved. For instance, an automated negotiation device which is not yet available needs to be mounted in
the EV that performs the enabling function required in the
transactive control. Price control is probably the most attractive
way for the commercial actor to regulate the charging behavior of
the EVs considering its easier implementation. It is especially
effective in the case of decreasing the charging in the peak time for
distribution system operator or the case of increasing the load for
transmission system operator. For the price control to become a
success, more research is needed in price responsive models or
1224
price elasticity models to obtain satisfactory performance and grid

reliability.
The following benets of present study can be identied:
The study outlines potential services that EV eet operator

could provide in a smart grid;
The advantage and disadvantage of centralized, transactive

control and price control are discussed, which gives a basis for
comparing available methods for future developments;
Details of the modeling method and algorithms of each control
strategy are illustrated by showing the key formulas and compared in term of their performance, calculation time, communication and computation requirements etc.
8.2. Recommendations on future research directions in the area

Based on the discussions in the present study, future research
directions are suggested as following:
Coordinate the multi-objectives of smart charging of EVs
Recently, the trend in smart charging of EVs is to integrate the

interests of EV owners, ancillary services required by the
transmission system operator and the distribution system
operator. This is because the conict might exist in certain
cases when involving electric vehicle exibility which is
explained in Section 2.2.5.
Real time EV eet management
It should be observed and emphasized that the above discussion did not consider the real time operations, i.e., there is no
continuous monitoring and assessment of the state of electric
vehicle battery dynamics and therefore lack of the appropriate
response in abnormal situations. This means that real time control
method considering the dynamic behavior of EV eet and power
systems should be developed as well.
References
[1] European Commission, Energy Roadmap 2050; 2011.
[2] Wang Z, Yang L. Delinking indicators on regional industry development and
carbon emissions: BeijingTianjinHebei economic band case. Ecol Indic
2015;48:418.
[3] Wang Z, Yin F, Zhang Y, Zhang X. An empirical research on the inuencing
factors of regional CO2 emissions: evidence from Beijing city, China. Appl
Energy 2012;100:27784.
[4] Blesl M, Das A, Fahl U, Remme U. Role of energy efciency standards in
reducing CO2 emissions in Germany: an assessment with TIMES. Energy
Policy 2007;35(2):77285.
[5] Wang Z, Wang C, Yin J. Strategies for addressing climate change on the
industrial level: affecting factors to CO2 emissions of energy-intensive
industries in China. Nat Hazards 2014;75(S2):S30317.
[6] Zhang B, Wang Z. Inter-rm collaborations on carbon emission reduction
within industrial chains in China: practices, drivers and effects on rms'
performances. Energy Econ 2014;42:11531.
[7] European Environment Agency, Greenhouse gas emission trends, 2013.
[8] Ellabban O, Abu-Rub H, Blaabjerg F. Renewable energy resources: current
status, future prospects and their enabling technology. Renew Sustain
Energy Rev 2014;39:74864.
[9] Paliwal P, Patidar NP, Nema RK. Planning of grid integrated distributed
generators: a review of technology, objectives and techniques. Renew Sustain Energy Rev 2014;40:55770.
[10] Sun Z, Li K, Yang Z, Niu Q, Foley A. Impact of electric vehicles on a carbon
constrained power systema post 2020 case study. J Power Energy Eng
2015;03(04):11422.
[11] Fadaeenejad M, Saberian AM, Fadaee M, Radzi MAM, Hizam H, AbKadir MZA.
The present and future of smart power grid in developing countries. Renew
Sustain Energy Rev 2014;29:82834.
[12] Gouveia C, Rua D, Soares FJ, Moreira C, Matos PG, Lopes JAP. Development
and implementation of Portuguese smart distribution system. Electr Power
Syst Res 2015;120:15062.
[13] Parks K, Denholm P, Markel AJ. Costs and emissions associated with plug-in
hybrid electric vehicle charging in the Xcel Energy Colorado service territory.
CO: National Renewable Energy Laboratory Golden; 2007.
[14] Sndergren C. Edison project: electric vehicles in future market models

2011:15.
[15] Bessa RJ, Matos MA. Economic and technical management of an aggregation
agent for electric vehicles: a literature survey. Eur Trans Electr Power
2012;22(3):33450.
[16] San Romn TG, Momber I, Abbad MR, Miralles Snchez. Regulatory framework and business models for charging plug-in electric vehicles: infrastructure, agents, and commercial relationships. Energy Policy 2011;39
(10):636075.
[17] Callaway DS, Hiskens IA. Achieving controllability of electric loads. Proc IEEE
2011;99(1):18499.
[18] Heussen K, You S, Biegel B, Hansen LH, Andersen KB. Indirect control for
demand side management a conceptual introduction. In: IEEE PES innovative smart grid technologies conference Europe; 2012. p. 18.
[19] Lyon TP, Michelin M, Jongejan A. Is smart charging policy for electric
vehicles worthwhile? Energy Policy 2012;41:25968.
[20] Garca-Villalobos J, Zamora I, San Martn JI, Asensio FJ, Aperribay V. Plug-in
electric vehicles in electric distribution networks: a review of smart charging
approaches. Renew Sustain Energy Rev 2014;38:71731.
[21] Mwasilu F, Justo JJ, Kim E-K, Do TD, Jung J-W. Electric vehicles and smart grid
interaction: a review on vehicle to grid and renewable energy sources
integration. Renew Sustain Energy Rev 2014;34:50116.
[22] Richardson DB. Electric vehicles and the electric grid: a review of modeling
approaches, impacts, and renewable energy integration. Renew Sustain
Energy Rev 2013;9:24754.
[23] Green II RC, Wang L, Alam M. The impact of plug-in hybrid electric vehicles
on distribution networks: a review and outlook. Renew Sustain Energy Rev
2011;15(1):54453.
[24] Yong JY, Ramachandaramurthy VK, Tan KM, Mithulananthan N. A review on
the state-of-the-art technologies of electric vehicle, its impacts and prospects. Renew Sustain Energy Rev 2015;49:36585.
[25] Yang Z, Li K, Foley A. Computational scheduling methods for integrating
plug-in electric vehicles with power systems: a review. Renew Sustain
Energy Rev 2015;51:396416.
[26] Lopes JAP, Soares FJ, Almeida PMR. Integration of electric vehicles in the
electric power system. Proc IEEE 2011;99(1):16883.
[27] EURELECTRIC. Ancillary services unbundling electricity products an
emerging market; 2004.
[28] Kempton W, Tomic J. Vehicle-to-grid power fundamentals: calculating
capacity and net revenue. J Power Sources 2005;144(1):26879.
[29] Kempton W, Tomic J. Vehicle-to-grid power implementation: from stabilizing the grid to supporting large-scale renewable energy. J Power Sources
2005;144(1):28094.
[30] Dallinger D, Krampe D, Wietschel M. Vehicle-to-grid regulation reserves
based on a dynamic simulation of mobility behavior. IEEE Trans Smart Grid
2011;2(2):30213.
[31] Rotering N, Ilic M. Optimal charge control of plug-in hybrid electric vehicles
in deregulated electricity markets. IEEE Trans Power Syst 2011;26(3):10219.
[32] Almeida PMR, Soares FJ, Lopes JAP. Electric vehicles contribution for frequency control with inertial emulation. Electr Power Syst Res 2015;127:141
50.
[33] Sortomme E, El-Sharkawi MA. Optimal scheduling of vehicle-to-grid energy
and ancillary services. IEEE Trans Smart Grid 2012;3(1):3519.
[34] Heydt GT. The impact of electric vehicle deployment on load management
strategies. IEEE Power Eng Rev 1983;PER-3(5):412.
[35] Rahman S, Shrestha GB. An investigation into the impact of electric vehicle
load on the electric utility distribution system. IEEE Trans Power Deliv
1993;8(2):5917.
[36] Sortomme E, Hindi M, MacPherson S, Venkata S. Coordinated charging of
plug-in hybrid electric vehicles to minimize distribution system losses. IEEE
Trans Smart Grid 2011;2(1):198205.
[37] Morais H, Sousa T, Vale Z, Faria P. Evaluation of the electric vehicle impact in
the power demand curve in a smart grid environment. Energy Convers
Manag 2014;82:26882.
[38] Wu C, Mohsenian-Rad H, Huang J, Jatskevich J. PEV-based combined frequency and voltage regulation for smart grid. In: Proceeding of the IEEE PES
innovative smart grid technologies (ISGT); 2012. p. 16.
[39] Leemput N, Geth F, Van Roy J, Bscher J, Driesen J. Reactive power support in
residential LV distribution grids through electric vehicle charging. Sustain
Energy Grids Netw 2015;3:2435.
[40] Sousa T, Morais H, Vale Z, Castro R. A multi-objective optimization of the
active and reactive resource scheduling at a distribution level in a smart grid
context. Energy 2015;85:23650.
[41] Li P, Degobert P, Robyns B, Francois B. Participation in the frequency regulation control of a resilient microgrid for a distribution network. Int J Integr
Energy Syst 2009;1(1).
[42] Rueda-Medina AC, Padilha-Feltrin A, Mantovani JRS. Active power reserve for
frequency control provided by distributed generators in distribution networks. In: Proceeding IEEE PES general meeting|conference & exposition;
2014. p. 15.
[43] Liu L, Kong F, Liu X, Peng Y, Wang Q. A review on electric vehicles interacting
with renewable energy in smart grid. Renew Sustain Energy Rev
2015;51:64861.
[44] Lund H, Kempton W. Integration of renewable energy into the transport and
electricity sectors through V2G. Energy Policy 2008;36(9):357887.
[45] Bellekom S, Benders R, Pelgrm S, Moll H. Electric cars and wind energy: two
problems, one solution? A study to combine wind energy and electric cars in
2020 in The Netherlands Energy 2012;45(1):85966.
[46] Dallinger D, Wietschel M. Grid integration of intermittent renewable energy
sources using price-responsive plug-in electric vehicles. Renew Sustain
Energy Rev 2012;16(5):337082.
[47] Saber AY, Venayagamoorthy GK. Plug-in vehicles and renewable energy
sources for cost and emission reductions. IEEE Trans Ind Electron 2011;58
(4):122938.
[48] Saber AY, Venayagamoorthy GK. Resource scheduling under uncertainty in a
smart grid with renewables and plug-in vehicles. IEEE Syst J 2012;6(1):103
9.
[49] Schuller A, Flath CM, Gottwalt S. Quantifying load exibility of electric
vehicles for renewable energy integration. Appl Energy 2015;151:33544.
[50] Birnie DP. Solar-to-vehicle (S2V) systems for powering commuters of the
future. J Power Sources 2009;186(2):53942.
[51] Zhang Q, Tezuka T, Ishihara KN, Mclellan BC. Integration of PV power into
future low-carbon smart electricity systems with EV and HP in Kansai Area,
Japan. Renew Energy 2012;44:99108.
[52] Hu J, You S, Lind M, stergaard J. Coordinated charging of electric vehicle for
congestion prevention in the distribution grid. IEEE Trans Smart Grids
2014;5(2):70311.
[53] Kristoffersen TK, Capion K, Meibom P. Optimal charging of electric drive
vehicles in a market environment. Appl Energy 2010;88:19408.
[54] Al-Awami AT, Sortomme E. Coordinating vehicle-to-grid services with
energy trading. IEEE Trans Smart Grid 2012;3(1):45362.
[55] Johannsson H, Hansen H, Hansen LH, Holm-Hansen H-H, Cajar P, Bindner
HW, Samuelsson O. Coordination of system needs and provision of services.
In: Proceedings of the 22nd international conference and exhibition on
electricity distribution (CIRED 2013); 2013. p. 14.
[56] Sbordone D, Bertini I, Di Pietra B, Falvo MC, Genovese A, Martirano L. EV fast
charging stations and energy storage technologies: a real implementation in
the smart micro grid paradigm. Electr Power Syst Res 2014.
[57] Guzzella L, Sciarretta A. Vehicle propulsion systems: introduction to modeling and optimization. Berlin, Heidelberg: Springer; 2007.
[58] Zhang C, Li K, Mcloone S, Yang Z. Battery modelling methods for electric
vehicles a review. In: Proceedings of the European control conference
(ECC); 2014. p. 267378.
[59] Blockx P, Van Den Bossche P, Omar N, Van Mierlo J, Rotthier B, Cappelle J.
Beyond the plug and socket: towards safe standardized charging infrastructures. In: Proceedings of the world electric vehicle symposium and
exhibition (EVS27); 2013. p. 110.
[60] Tang WW, Wu YM, Qin J. Comparative study on electric vehicle charging
standards at home and abroad. Adv Mater Res 2013;608609:15539.
[61] Bakker S. Standardization of EV recharging infrastructures; 2013.
[62] Ka bisch S, Schmitt A, Winter M, Heuer J. Interconnections and communications of electric vehicles and smart grids. In: Proceedings of the rst IEEE
international conference on smart grid communications (SmartGridComm),
IEEE; 2010. p. 161166.
[63] Hansch K, Pelzer A, Komarnicki P, Groning S, Schmutzler J, Wietfeld C, Heuer
J, Muller R. An ISO/IEC 15118 conformance testing system architecture. In:
Proceedings of the IEEE PES general meeting|conference & exposition; 2014.
p. 15.
[64] Yang Y, Roy S. Grouping-based MAC protocols for EV charging data transmission in smart metering network. IEEE J Sel Areas Commun 2014;32
(7):132843.
[65] American National Standards Institute. Standardization roadmap for electric
vehicles; 2013.
[66] Mltin M, Gitte C, Schmeck H. Smart grid-ready communication protocols
and services for a customer-friendly electromobility experience. In: GI-Jahrestagung; 2013. p. 147084.
[67] Shahidinejad S, Bibeau E, Filizadeh S. Statistical development of a duty cycle
for plug-in vehicles in a North American urban setting using eet information. IEEE Trans Veh Technol 2010;59(8):37109.
[68] Soares J, Canizes B, Lobo C, Vale Z, Morais H. Electric vehicle scenario
simulator tool for smart grid operators. Energies 2012;5(12):188199.
[69] Azadfar E, Sreeram V, Harries D. The investigation of the major factors
inuencing plug-in electric vehicle driving patterns and charging behaviour.
Renew Sustain Energy Rev 2015;42:106576.
[70] Rolim CC, Gonalves GN, Farias TL, Rodrigues . Impacts of electric vehicle
adoption on driver behavior and environmental performance. Procedia - Soc
Behav Sci 2012;54:70615.
[71] Ipakchi A. Demand side and distributed resource managementa transactive
solution. In: Proceedings of the power and energy society general meeting,
IEEE; 2011. p. 18.
[72] Melton RB. GridWise transactive energy framework (DRAFT Version). Pacic
Northwest National Laboratory (PNNL), Richland, WA (US); 2013.
[73] Kok JK, Warmer CJ, Kamphuis IG. PowerMatcher: multiagent control in the
electricity infrastructure. In: Proceedings of the fourth international joint
conference on autonomous agents and multiagent systems; 2005. p. 7582.
[74] Shao S, Zhang T, Pipattanasomporn M, Rahman S. Impact of TOU rates on
distribution load shapes in a smart grid with PHEV penetration. In: Proceeedings of the IEEE PES transmission and distribution conference and
exposition: smart solutions for a changing world; 2010.
[75] Yu R, Yang W, Rahardja S. A statistical demand-price model with its application in optimal real-time price. IEEE Trans Smart Grid 2012;3(4):173442.
1225
[76] Richardson P, Flynn D, Keane A. Local versus centralized charging strategies

for electric vehicles in low voltage distribution systems. IEEE Trans Smart
Grid 2012;3(2):10208.
[77] Karfopoulos EL, Hatziargyriou ND. A multi-agent system for controlled
charging of a large population of electric vehicles. IEEE Trans Power Syst
2013;28(2):1196204.
[78] Sundstrom O, Binding C. Flexible charging optimization for electric vehicles
considering distribution grid constraints. IEEE Trans Smart Grid 2012;3
(1):2637.
[79] Iqbal M, Azam M, Naeem M, Khwaja AS, Anpalagan A. Optimization classication, algorithms and tools for renewable energy: a review. Renew Sustain
Energy Rev 2014;39:64054.
[80] Clement-Nyns K, Haesen E, Driesen J. The impact of charging plug-in hybrid
electric vehicles on a residential distribution grid. IEEE Trans Power Syst
2010;25(1):37180.
[81] Reza Norouzi M, Ahmadi A, Esmaeel Nezhad A, Ghaedi A. Mixed integer
programming of multi-objective security-constrained hydro/thermal unit
commitment. Renew Sustain Energy Rev 2014;29:91123.
[82] Sousa T, Soares J, Vale Z, Morais H. Day-ahead resource scheduling in smart
grids considering vehicle-to-grid and network constraints. Appl Energy
2012;96:18393.
[83] Wang J, Liu C, Ton D, Zhou Y, Kim J, Vyas A. Impact of plug-in hybrid electric
vehicles on power systems with demand response and wind power. Energy
Policy 2011;39(7):401621.
[84] Franco JF, Rider MJ, Romero R. A mixed-integer linear programming model
for the electric vehicle charging coordination problem in unbalanced electrical distribution systems. IEEE Trans Smart Grid 2015;PP(99) 11.
[85] Klotz E, Newman AM. Practical guidelines for solving difcult mixed integer
linear programs. Surv Oper Res Manag Sci 2013;18(12):1832.
[86] Sousa T, Vale Z, Carvalho JP, Pinto T, Morais H. A hybrid simulated annealing
approach to handle energy resource management considering an intensive
use of electric vehicles. Energy 2014;67:8196.
[87] Balram P, Tuan LA, Tjernberg LB. Modeling of regulating power market based
on AC optimal power ow considering losses and electric vehicles. In: Proceedings of the IEEE innovative smart grid technologies-Asia (ISGT Asia);
2013. p. 16.
[88] Deilami S, Masoum AS, Moses PS, Masoum MAS. Real-time coordination of
plug-in electric vehicle charging in smart grids to minimize power losses
and improve voltage prole. IEEE Trans Smart Grid 2011;2(3):45667.
[89] Ansari M, Al-Awami AT, Sortomme E, Abidoeric MA. Coordinated bidding of
ancillary services for vehicle-to-grid using fuzzy optimization. IEEE Trans
Smart Grid 2015;6(1):26170.
[90] Linderoth J, Shapiro A, Wright S. The empirical behavior of sampling methods for stochastic programming. Ann Oper Res 2006;142(1):21541.
[91] Eminoglu U, Hocaoglu MH. Distribution systems forward/backward sweepbased power ow algorithms: a review and comparison study. Electr Power
Compon Syst 2008;37(1):91110.
[92] Fluhr J, Ahlert KH, Weinhardt C. A stochastic model for simulating the
availability of electric vehicles for services to the power grid. In: hicss; 2010.
p. 110.
[93] Su W, Chow MY. Evaluation on intelligent energy management system for
PHEVs/PEVs using Monte Carlo method. In: 4th international conference on
electric utility deregulation and restructuring and power technologies
(DRPT); 2011. p. 167580.
[94] Ben-Tal A, El Ghaoui L, Nemirovski A. Robust optimization. Princeton series
in applied mathematics; 2009.
[95] Soyster A. Technical noteconvex programming with set-inclusive constraints and applications to inexact linear programming. Oper Res
1973;21:11547.
[96] Zheng T, Zhao J, Litvinov E, Zhao F. Robust optimization and its application to
power system operation. In: Proceedings CIGR SC C2 Sess; 2012.
[97] Bertsimas D, Litvinov E, Sun XA, Zhao J, Zheng T. Adaptive robust optimization for the security constrained unit commitment problem. IEEE Trans
Power Syst 2012;28(1):5263.
[98] Hajimiragha AH, Canizares CA, Fowler MW, Moazeni S, Elkamel A. A robust
optimization approach for planning the transition to plug-in hybrid electric
vehicles. IEEE Trans Power Syst 2011;26(4):226474.
[99] Battistelli C, Baringo L, Conejo AJ. Optimal energy management of small
electric energy systems including V2G facilities and renewable energy
sources. Electr Power Syst Res 2012;92:509.
[100] Soares J, Sousa T, Morais H, Vale Z, Canizes B, Silva A. Application-specic
modied particle swarm optimization for energy resource scheduling considering vehicle-to-grid. Appl Soft Comput 2013;13(11):426480.
[101] Johnson D, Garey M. Computers and intractability: a guide to the theory of
NP-completeness. New York, USA: W.H. Freeman; 1979.
[102] Baos R, Manzano-Agugliaro F, Montoya FG, Gil C, Alcayde A, Gmez J.
Optimization methods applied to renewable and sustainable energy: a
review. Renew Sustain Energy Rev 2011;15(4):175366.
[103] Beheshti Z, Shamsudding S. A review of population-based meta-heuristic
algorithms. Int J Adv Soft Comput Appl 2013;5:135.
[104] Lee K, El-Sharkawi M. Modern heuristic optimization techniques: theory and
applications to power systems; 2008.
[105] Govardhan M, Roy R. Generation scheduling in smart grid environment using
global best articial bee colony algorithm. Int J Electr Power Energy Syst
2015;64:26074.
1226
[106] El-Zonkoly A. Intelligent energy management of optimally located renewable

energy systems incorporating PHEV. Energy Convers Manag 2014;84:42735.
[107] Grunewald A, Hardt S, Mielke M, Bruck R. A decentralized charge management for electric vehicles using a genetic algorithm: case study and proof-ofconcept in Java and FPGA. In: Proceedings of the IEEE congress on evolutionary computation; 2012. p. 17.
[108] Bashash S, Moura SJ, Forman JC, Fathy HK. Plug-in hybrid electric vehicle
charge pattern optimization for energy cost and battery longevity. J Power
Sources 2011;196(1):5419.
[109] Jewell N, Bai L, Naber J, McIntyre ML. Analysis of electric vehicle charge
scheduling and effects on electricity demand costs. Energy Syst 2013;5
(4):76786.
[110] Sousa T, Morais H, Vale Z, Faria P, Soares J. Intelligent energy resource
management considering vehicle-to-grid: a simulated annealing approach.
IEEE Trans Smart Grid 2012;3(1):53542.
[111] Sousa T, Morais H, Morais R, Vale Z. A new heuristic providing an effective
initial solution for a simulated annealing approach to energy resource scheduling in smart grids. In: Proceedings of the IEEE symposium on computational intelligence applications in smart grid (CIASG); 2014. p. 18.
[112] Ghanbarzadeh T, Goleijani S, Moghaddam MP. Reliability constrained unit
commitment with electric vehicle to grid using hybrid particle swarm optimization and ant colony optimization. In: Proceedings of the IEEE power and
energy society general meeting; 2011. p. 17.
[113] Di Giorgio A, Liberati F, Canale S. Electric vehicles charging control in a smart
grid: a model predictive control approach. Control Eng Pract 2014;22:14762.
[114] Su W, Wang J, Zhang K, Huang AQ. Model predictive control-based power
dispatch for distribution system considering plug-in electric vehicle uncertainty. Electr Power Syst Res 2014;106:2935.
[115] Halvgaard R, Poulsen NK, Madsen H, Jorgensen JB, Marra F, Bondy DEM.
Electric vehicle charge planning using economic model predictive control. In:
Proceedings of the 2012 IEEE international electric vehicle conference; 2012.
p. 16.
[116] Bansal S, Zeilinger MN, Tomlin CJ. Plug-and-play model predictive control for
electric vehicle charging and voltage control in smart grids. In: Proceedings
of the 53rd IEEE conference on decision and control; 2014. p. 5894900.
[117] Zuccaro L, Di Giorgio A, Liberati F, Canale S, Lanna A, Pallares VF, Blanco AM,
Escobar RU, Ratej J, Mehle B, Mehle U. Smart vehicle to grid interface project:
[118]
[119]
[120]
[121]
[122]
[123]
[124]
[125]
[126]
[127]
[128]
electromobility management system architecture and eld test results. In:

Proceedings of the IEEE International Electric Vehicle Conference (IEVC);
2014. p. 17.
Hu J, Yang G, Bindner HW. Network constrained transactive control for
electric vehicles integration. In: Proceedings of the IEEE power and energy
society general meeting; 2015.
Ma Z, Callaway DS, Hiskens IA. Decentralized charging control of large
populations of plug-in electric vehicles. IEEE Trans Control Syst Technol
2013;21(1):6778.
Gan L, Topcu U, Low S. Optimal decentralized protocol for electric vehicle
charging. IEEE Trans Power Syst 2013;28(2):94051.
Vandael S, Claessens B, Hommelberg M, Holvoet T, Deconinck G. A scalable
three-step approach for demand side management of plug-in hybrid vehicles. IEEE Trans Smart Grid 2013;4(2):7208.
Weckx S, DHulst R, Driesen J. Primary and secondary frequency support by a
multi-agent demand control system. Power Syst IEEE Trans 2014;PP(99):1
11.
Faruqui A, Hledik R, Levy A, Madian A. Will smart prices induce smart
charging of electric vehicles? Brattle Gr. Discuss; 2011.
Huang S, Wu Q, Oren SS, Li R, Liu Z. Distribution locational marginal pricing
through quadratic programming for congestion management in distribution
networks. IEEE Trans Power Syst 2015;30(4):21708.
Sorrell S, Dimitropoulos J. The rebound effect: microeconomic denitions,
limitations and extensions. Ecol Econ 2008;65(3):63649.
Wang Z, Lu M, Wang J-C. Direct rebound effect on urban residential electricity use: an empirical study in China. Renew Sustain Energy Rev 2014;30
(0):12432.
Han X, Bindner HW, Hansen H, Rasmussen KS, Cajar PD, Ding Y, Tackie DV.
Identication of kick-back from normal situations due to activation of exible
demand. Danish iPower Consortium, Copenhagen; 2014.
Li R, Wu Q, Oren SS. Distribution locational marginal pricing for optimal
electric vehicle charging management. IEEE Trans Power Syst 2014;29
(1):20311.
IEEE Vehicle Power and Propulsion Conference (VPPC), September 3-5, 2008, Harbin, China
Integrated Vehicle Dynamics Control

State-of-the Art Review
Fan Yu* Dao-Fei Li** D.A. Crolla***
*Shanghai Jiao Tong University, Shanghai, 200240, China. fanyu@sjtu.edu.cn
**Zhejiang University, Hangzhou, 310027, China. duffayli@gmail.com
***University of Sunderland, Sunderland SR6 0DD, UK. dave.crolla@sunderland.ac.uk
AbstractChassis
control
systems
have
evolved
dramatically over the past two decades and their impacts on
vehicle dynamics can be usefully separated into the three
directions, i.e. lateral, longitudinal and vertical directions.
Accordingly, the state survey of chassis control systems can
be reviewed in following sub-system areas, i.e. steering,
driveline/braking and suspension. The developments within
each of these areas have progressed at different rates and
each has had different impacts on improving vehicle
behavior in relation to safety, ride, handling dynamics or
economy. However, the biggest challenge is in the whole
chassis integration of these sub-systems to avoid their
interventions and thus to improve overall vehicle dynamics
performance. Hence, a hot research topic, named Integrated
Vehicle Dynamics Control (IVDC) or Integrated Chassis
Control, arose. Based on the published literatures in the
recent twenty years, this paper presents a comprehensive
state of the art survey on IVDC. First, the roadmap and
methodologies of IVDC are reviewed, and then the control
strategies of coordination between the subsystems are
summarized. At present, integration technique between
steering and braking/traction has been most concerned, and
is being researched and developed intensively both in
academic and industrial aspects. It can be expected that
once X-by-Wire technology and actuator hardware are
further developed, more potential benefits of IVDC can be
obtained.
KeywordsVehicle Dynamics; Integrated Chassis Control;
Active Steering; Electronic Stability Control; Active
Suspension; Review
I. INTRODUCTION AND BACKGROUND

In order to cope with the complicated operation
conditions and to improve vehicle safety and comfort
meanwhile, various active control systems (such as ABS,
4WS, ESP and semi-active/active suspensions, etc.) were
equipped in vehicles one after another since the late
1970s. Generally, these control systems are developed
and supplied by different manufacturers. But when they
work together, many problems could occur since the
coupling of vehicle dynamics inherently exist. Therefore,
since the mid-1980s, the research on integrated vehicle
dynamics control has become a focus and attracted much
______________________
This work was sponsored by Nature Science Foundation of China (Grant

No: 50575141)
C 2008 IEEE.
978-1-4244-1849-7/08/$25.00
attention.
The co-existence of several control subsystems can
cause two main problems. First, the design of software
and hardware will become more complicated due to the
dramatically increased number of sensors and signal
cables. Second, because of the possible function
overlapping among these systems, their conflicts of
control objectives and actions will inevitably occur. If
without coordination, the dynamic performance of these
systems could be worse than that of individual system or
even worse than a passive system without any active
control. For example, ABS and ESP are both based on the
tire slip ratio control, however the primary objective of
ABS is to maintain the slip ratio around the optimal value
(normally corresponding to peak friction coefficient),
while the objective of ESP is to control wheel slip ratios
properly for vehicle stability improvement via braking the
properly-selected wheels.
Therefore, the two key problems of integrated vehicle
dynamics control to be solved are: 1) to avoid the
conflicts and interventions among different subsystems;
2) to exploit the potentials of each subsystems by
communication and coordination among the sub-systems
since different system has different action domain.
In practice, vehicle electric control systems could cover
much broader domain, including body electric control
systems, but this paper is mainly to review the Integrated
Vehicle Dynamics Control (IVDC), also called
Integrated Chassis Control based on many literatures
published within recent 20 years. Although many relevant
researches have been published recently, the definition of
IVDC is still not very definitive. However, its attributes is
obvious, i.e. it needs to coordinate two or more
subsystems systematically according to control objectives
and actions in terms of both software and hardware,
rather than simply put the subsystems altogether.
Besides the obvious benefits of IVDC, including
reduction of number of sensors and actuators, less
mounting space and weight, etc., IVDC can further
exploit the potential benefits of the control systems and
thus enhance overall vehicle dynamic performances. By
integrating different subsystems, the comprehensive
performance of IVDC can be much better with the
potential benefits as shown in Figure 1, in which the
sphere diameter can be maximized by system
coordination. Hence, overall vehicle dynamics
performances, in directions of longitudinal, lateral and
vertical, can be improved.
Integrated Safety Technology is being developed which

combines the active and passive safety systems.
III. INTEGRATED CONTROL STRUCTURE
A. Decentralized Control Structure
Fig 1.
Potential benefits via system integration
II. DEVELOPMENT OF INTEGRATED VEHICLE DYNAMICS

CONTROL
The period between 1980s and 1990s is the beginning
stage of IVDC. Most tentative researches were first
carried out by Japanese manufacturers. In 1985 Tokyo
International Auto Exhibition, Nissan presented its
concept car ARC-X, in which the subsystems were able
to achieve some integrated functions by communication
and coordination [1]. In 1987, the concept car FXV-II
presented by Toyota, integrated active air suspensions,
4WS, engine control, gearbox ratio control, 4WD and
ABS [2]. In 1991, Toyota Sorarer model coordinated
4WS, ABS and TCS[3, 4]. In 1992 AVEC [5], Prof.
Wallentowiz pointed out that an integrated control system
should include three parts: the hardware integration,
function integration and research integration.
Since mid 1990s when ESP was developed, these
recent ten years is the rapid developing period of IVDC.
Many achievements have been made both in academic
and industry fields. In 2002, GM adopted Delphis
Integrated Chassis Control System in its Cadillac Seville,
in which the stability system based on braking and
continuous adjustable-damper are integratedly designed.
The good ride comfort performance can be achieved in
normal circumstances, while enhanced handling
performance can be obtained in emergency cases by
sacrificing ride comfort. Besides its original differential
braking and engine interventions, Continental further
introduced Active Front Steering to its second generation
ESP[6,7]. In 2006, Toyota adopted the comprehensive
safety concept on its Lexus 460. In the hardware aspect,
its control units are classified into four ECU function
groups (i.e. power transmission control, safety control,
body control and multi-media control). In the software
aspect, Lexus 460 uses integrated vehicle dynamic
management system to manage the ABS, traction control,
active braking intervention and steering system
congruously.
Recently, the integrated chassis control is gradually
extending to whole vehicle control, even including
driver-vehicle-road close-loop integrated control system.
All these subsystems will be covered in a single frame,
including Driving Assist System (e.g. cruising control),
active safety (e.g. ESC) and passive safety system (e.g.
supplemental restraint system). Automotive suppliers,
such as Delphi, Bosch, Continental and TRW have made
efforts to achieve this goal. Currently, a so-called
According to the differences in design strategies for

integrated control, the relevant literatures may be
classified into two categories: in the fashions of
bottom-up and top-down respectively. The former
normally adopts a decentralized control structure, while
the latter uses a centralized control structure. A typical
decentralized control structure is adopted in bottom-up
fashion shown in Figure 2. But for more complicated
integrated systems, some researchers currently tend to
adopt a multi-layer control structure to cope with the
coordination and distribution of subsystems.
Fig 2. Decentralized control structure
In the decentralized control structure (as shown in

Figure 2), each subsystem is still independent to some
extent, but it can cooperate with other subsystems by bus
systems to achieve a certain function if necessary. Such
integration is very limited, because it just integrates the
sensors and related hardwares. Moreover, such
integrations have to be developed by OEM and thus
automotive suppliers only need to provide corresponding
interfaces necessary to their subsystems. Such
decentralized structure is not very suitable for vehicle
integrated control, as pointed out by Professor Gordon in
literature [8], though it can achieve some effects of
integration by redesigning the controller of each
subsystem (e.g. taking the information of other
subsystems into account). Practically, this structure was
adobted in most early researches on IVDC. For instance,
some researches carried out by Mitsubishi and Toyota
used the decentralized structure [3, 9]. In addition,
compared to braking and steering systems, suspension
systems aiming to improve ride comfort can only
indirectly affect the handling performance. This might be
the reason why decentralized structure has been used by
some researchers [43-44] for integrating suspension
system with other subsystems.
B. Centralized Control Structure
With centralized control structure, as shown in Figure
3, a global controller offers control inputs to all the
subsystems. Unlike decentralized control structure, this

kind of IVDC controller can only be co-developed by
OEM and suppliers. Due to the top-down strategy, the
integrated control system can be developed more
comprehensively by taking the whole systems into
account and thus the centralized control structure can
provide better performance. However, the more
sophisticated the whole system is, the more difficult the
design and development can be. Considering the global
controllers computing load, such structure requires that
the ECU be of much more powerful. Also, such
developed control system may become less flexible, since
when a new subsystem is added it is inevitable to
redesign the system. For these reasons, centralized control
structure might not be available for integrated vehicle
dynamics control yet so far, but it is a good alternative for
those subsystems that are closely related and coupled. In
the aspect of difficulty of controller design, the method of
using multivariable control stated in literature [10] might
be very difficult, but by dividing the system into several
control loops and decoupling the tire dynamics
characteristics, it is able to take the non-linear
characteristics of tire into account directly and thus the
design difficulty can be greatly reduced [11, 12].
Fig 3. Fully centralized control structure
C. Multi-layer Control Structure

Considering the comprehensive advantages in sense of
both software and hardware, a more suitable structure
scheme, the so-called Multi-layer Control Structure, is
currently adopted, as shown in Figure 4. When using this
structure, suppliers are still in charge of the design of the
subsystems, while OEMs are responsible for the
integration, i.e. vehicle dynamics coordination
controller (or upper controller). Coordination controller
can have two main functions. First, according to the
dynamic state of the vehicle, the controller determines the
desired control forces or moments (e.g. active stability
yaw moment Mza), then distributes them to the desired
state values of each subsystems (e.g. tire slip ratio or tire
side slip angle). Then the controller supplies the actuation
layer controllers with these desired control forces and
moments to track or regulate. Second, the coordination
controller can shift the subsystems to work at different
modes (e.g. to switch the suspension strategy between tire
gripping or ride comfort strategies).
Fig. 4 Multi-layer control structure
IV. COORDINATION AND INTEGRATION STRATEGY

The difficult task for control system integration
strategy is how to distribute and coordinate several
actuators properly and make them work together
harmoniously. This paper mainly focuses on longitudinal
and lateral control system integration, and discusses
suspension system integration.
A. Longitudinal and Lateral System Integration
The major systems of longitudinal and lateral dynamics
control are the braking/traction and steering systems,
especially referring to the integration of ABS, ESP and
AFS/4WS.
Longitudinal and lateral control systems have
advantages of both control effect and intervention
smoothness. First, when the lateral acceleration is small
enough, the side slip angle is small within the linear
region, thus good handling performance can be obtained
by steering system (e.g. 4WS and AFS). However, in
emergency cases (e.g. large lateral acceleration ay and
large side slip angle) non-linear tire characteristics will
exhibit. The desired control effect cannot be obtained
only by steering control, while DYC (e.g. ESP based on
braking system and AYC based on traction forces
distribution) can significantly enhance the handling
performance. Second, as for intervention smoothness,
when the control system exerts a braking intervention,
ESP could cause a sudden longitudinal velocity variation
and thus reduces the comfort performance to some extent.
On the contrary, AYC redistributes the traction forces on
both sides (the left and right), ensuring the total
longitudinal force un-affected, hence the intervention of
the control system become relatively more smooth. When
lateral acceleration is relatively small, active steering can
improve the vehicle yaw performance with less
interruption to the driver.
Following review some typical literatures related
researches according to different integration extents and
control structures.
Decentralized Structure: to coordinate steering and
braking/traction subsystems by information interaction
Taheri and Law took the tire lateral force into account
in the design of ABS controller [13]. In steering/braking
circumstances, by modifying front wheel steering angle
and the desired tire slip ratio and reducing rear wheels
desired slip ratio, such design can increase lateral forces

margin. In [14], ESP and EPS were coordinated as
follows: tire lateral adhesion limits were firstly estimated
from the information obtained by EPS, and then the
estimated information was sent to ESP controller to avoid
tire force saturation sent by reducing tire longitudinal
force when the tire characteristics entered the non-linear
region.
Multi-layer Structure: firstly generate an active yaw
moment then allocate it to subsystems
Vehicle lateral stability control in multi-layer control
structure can be mainly divided into two steps. The first is
to calculate the desired vehicle stability control
force/moment by using feedback controller. And then the
second is to distribute the control force/moment to each
subsystem accordingly.
As mentioned above, ESPs intervention can result in
the deceleration of the vehicle, which can cause a sudden
unpleasant impact to the driver. On the other hand, the
intervention of active steering is relatively smooth, thus
the energy consumed by controller can be lower.
Therefore, the general strategies of stability yaw moment
distribution between active steering and differential
braking are: 1) to give priority to adopt active steering
(front/rear wheel). 2) once active steering cannot generate
the desired stability yaw moment Mza, to use the
intervention of ESP or co-intervention of ESP and active
steering [15-17]. Generally, in practical account, it is first
to determine the effective working area of each
subsystems (e.g. according to body side slip angle [18]
or lateral acceleration [17] to determine the AFS and ESP
working areas), then distribute the desired Mza based on
empirical rules (e.g. fuzzy logic [17]).
Centralized Structure: comprehensive consideration
based on multivariable control method
Most literatures on the study of braking/steering
control integration adopt the centralized structure,
because the integration extent of such structure is the
largest. The top-down strategy can easily ensure good
control performances. However, a more powerful
hardware controller must be provided for extremely high
computation demand to meet the requirements of the
centralized structure.
Most researches are based on the assumption that tire
and vehicle dynamics are linear. First, the MIMO vehicle
dynamic state space equations are given as follow, then
using linear optimization control [10, 19-22], H
optimization control, robust eigenvalue assignment [23],
synthesis theory [24] or other linear feedback control
methods.

X = = A + B r + H f
r
M za
r
Other researchers linearized the non-linear vehicle
model under some certain working conditions, then
designed the controller based on linear model by using
robust theory [20, 24]. This method dealt with the
non-linear factors as uncertainties before modeling by
presetting the control gain to counteract the undesired
influence on the close loop system imposed by the

uncertainties.
Nagai carried out many studies on the integration of
active steering and DYC, combining the feed forward
control and LQR-based feedback control [10, 21, 22].
Afterwards, some other researchers furthered the research
on the design of feedback control law, for example using
robust control [25], fuzzy logic control [26] and robust
eigenvalue assignment [23] and so on. The common
features of these researches is that taking the active
steering angle f and stability yaw moment Mza as two
independent control inputs, i.e. assuming Mza can be
obtained only by the differential value between the left
and right sides tire longitudinal forces, while ignoring that
the yaw moment generated by steering. In addition, such
researches did not consider thoroughly the distribution of
Mza, but just simply obtained by the braking forces on the
left and right sides in one axle. Considering the
importance of tire in the dynamics control and the lateral
load transfer under large acceleration circumstances, such
method cannot ensure to generate a desired Mza.
Furthermore, Abe [27], Hattori [28] and the authors
research group further the study on how to distribute tire
forces by optimization. Unlike using linear tire model,
literatures [11, 12, 29~31] divided the controller into two
control loops (i.e. main loop and servo loop), taking the
tire as a force generator and putting it into the servo
loop to be dealt with. In this way, the tire non-linearities
can be directly taken into account. Tire forces can be
optimally distributed by using unconstrained and
constrained optimization with the consideration of road
adhesion, limits of actuators, etc. For instance, in
literature[12], limits of the actuators and adhesion of
active steering, braking and traction systems can be
converted to the upper and lower bounds of the tire
control force. Then the optimized distribution of tire force
can be realized by using control allocation, e.g. quadratic
program approach.
B. Suspension Integration
The main objective of suspension control is to enhance
the ride comfort. Because of the coupling of tire force, the
suspension system also couples the longitudinal and
lateral control systems. When integrating the suspension
into dynamic control, the main objective might still be to
reduce the intervention of lateral and longitudinal
systems. But in emergency cases, it is necessary to
sacrifice the ride comfort to ensure the performance of
ESP and AFS. Hence, the strategies of integration control
including suspension system can be divided into: 1) to
ensure tire grip and the fully usage of road adhesion by
using suspension to control tire vertical loads; 2) to
regulate the tire vertical forces and indirectly generate
assistant stability yaw moment through suspension
control (e.g. active suspension, CDC shock absorber or
active stabilizer bar, etc.). By doing so, the intervention of
active braking and the sudden change of longitudinal
velocity caused by braking can be minimized. The
relevant literatures are summarized as below.
(1) Enhance tire grip by active suspension to assist
longitudinal and lateral control system
Such researches include literatures [32~37], of which

the main principle is to design two strategies for active
and
semi-active
suspensions.
Under
general
circumstances, ride comfort strategy is adopted to
improve vertical dynamics; but in emergency, the active
suspension law will be switched to tire grip strategy by a
supervisor controller, temporarily sacrificing ride comfort
to ensure the braking and stability performance of ABS
and ESP. Literature [33] discussed the tire grip control
law of active suspension, making tire load and slip ratio
reach the maximum value simultaneously, in order to
make full use of the road adhesion. Literature [34] used
back-stepping method to study the related issue. March
and Shim designed the active front wheel steering and
suspension controller using fuzzy theory. The main
objective of the integrated suspension is still for roll
moment control, whilst assisting to improve the tire
lateral performance [37].
(2) Active suspension intervene yaw moment control
Smakman [38] integrated the active suspension and
active braking intervention using the similar strategy for
active steering and ESP integration, i.e. giving priority to
vertical load control of active suspension and the active
braking is enabled only when vertical load control cannot
adequately generate the stability yaw moment Mza.
Similarly, Hac [39], Kou [40] and others studied on the
integration of CDC damper and ESP system, and pointed
out that the acting time of active braking can be reduced
by using CDC suspension to generate partial Mza.
Active anti-roll bar or active roll torque distribution
control can affect the vehicle understeering characteristics
indirectly by changing the lateral transfer magnitude of
tire load between front and rear axles. Cooper and
Professor Crolla added a central differential controller in
the integration of active roll moment and active traction
moment distribution to allocate more traction moment to
the wheels with larger vertical load [41]. And they
concluded that the roll moment distribution between front
and rear axles can only affect the understeering
characteristics slightly. When lateral acceleration is small,
the total load transfer is relatively small and thus the
contribution of active suspension to yaw stability is
limited.
The two types of researches mentioned above aim to
develop longitudinal and lateral dynamics by further
integrating suspension control, while other researchers
attempted to improve vehicle vertical dynamics
performance. To reduce the rollover tendency under large
lateral acceleration circumstance, Lee [42] found that
inverse steering of rear wheel and increased front axle
suspension damper can assist the active front axel
anti-roll bar to increase the anti-roll moment so as to
avoid the rollover. Chen and Wang [43, 44] integrated the
suspension and steering control system based on the
output feedback random sub-optimal control to improve
steering quality and ride comfort.
longitudinal system), aiming to improve handling

dynamics, is still one of the main research topic in vehicle
dynamics and control area. The primary objective of
further integrating suspension control is to avoid the
negative intervention acting on longitudinal and lateral
systems to ensure best tire grip performance for each tire.
The development trend of vehicle control system is
X-by-wire, which eliminates the mechanical connection
between handling system and servo system, and the
energy and control information needed by servo system
will be transferred by wire. Besides the four conventional
active control subsystems (i.e. traction/braking, steering
and suspension), other control systems like camber, toe-in
and aerodynamics control have appeared in recent years.
With more practical applications, these control systems
may provide more design flexibility for vehicle dynamics
control and integration.
In the future, the theoretical study on integrated control
will probably focus on the following issues:
(1) More precise models of both vehicle and tire to be

taken to get higher control precision, including the
consideration of vehicle dynamics coupling and tire
non-linearities.
(2) Further investigation on the details of distribution

optimization of tire forces and moments for vehicle
stability control under different circumstances. For
example, the consideration of the limits of tire force
under various complicated conditions, which
includes vertical load, adhesion conditions and servo
system situations.
(3) More widely integrated control incorporated with

latest technology including active camber, toe-in and
X-by-wire techniques, etc.
REFERENCES
[1] Tokuda T. Cars in the 90s as a humanware. SAE Paper 885049,
1988
[2] Kizu R., Harada H., Minabe H. Electronic control of car chassis
[3]
[4]
[5]
[6]
[7]
[8]
[9]
V. CONCLUSIONS AND PROSPECTS
[10]
The active safety system integration (mainly referring

to the integration of active steering system and
[11]
present status and future perspective. International Congress on

Transportation Electronics, Convergence88, Detroit, USA, 1988,
pp. 173 188
Sato S., Inoue H., Tabata M. Integrated chassis control system for
improved vehicle dynamics. Proc. AVEC92, Yokohama, Japan,
1992, pp. 413-418
Tanaka H., Inoue H., Iwata H. Development of a vehicle
integrated control system. SAE Paper 925049, 1992
Roppenecher G., Wallentowitz H. Integration of chassis and
traction control systems: What is possible - what makes sense what is under development. J. Vehicle System Dynamics, Vol.
22, 1993, pp. 283-298
Chen Z. Status and tendency of the vehicle chasses control
technology. Automotive Engineering, CSAE, Vol. 28, No.2, pp.
105-113 (in Chinese)
Rieth P.E., Schwarz R. ESC II-ESC with active steering
intervention. SAE Paper 2004-01-0260, 2004
Gordon T., Howell M., Brandao F. Integrated control
methodologies for road vehicles. J. Vehicle System Dynamics.
Vol. 40, No.1, 2003, pp. 157-190
Mitamura R., Tani M., Tanaka T. System integration for new
mobility. SAE Paper 881773, 1988
Nagai M., Shino M., Gao F. Study on integrated control of active
front steer angle and direct yaw moment. JSAE Review, Vol.23,
No.3, 2002, pp. 309-315
Shen X., Li D., Yu F. Study on vehicle chassis control integration
based on general actuator-plant structure. Proc. AVEC06, Taipei,
Taiwan, 2006
[12] Li D., Yu F. A novel integrated vehicle chassis controller

[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
coordinating direct yaw moment control and active steering. SAE

Paper 2007-01-3642, 2007
Taheri S., Law E. Investigation of combined slip control braking
and closed loop four wheel steering systems for automobiles
during combined hard braking and severe steering. Proc.
ACC1990, San Diego, USA, 1990
Yasui Y., Kodama H., Momiyama M. Electronic stability control
(ESC) coordinated with electric power steering (EPS). FISITA
Congress 2006, Yokohama, Japan, 2006
Bedner E., Chen H. A supervisory control to manage brakes and
four-wheel-steer systems. SAE Paper 2004-01-1059, 2004
Burgio G., Zegelaar P. Integrated vehicle control using steering
and brakes. Int. J. Control, Vol. 79, No.5, 2006, pp.534~541
Goodarzi A., Alirezaie M. A new fuzzy-optimal integrated
AFS/DYC control strategy. Proc. AVEC06, Taipei, Taiwan, 2006
Park I., Park K. A study on integrated control of AFS and ESP for
the improvement of vehicle handling performance. Proc. of
Spring Conference 2005 of the Korean Society of Precision
Engineering, Pusan, Korean, 2005
Matsumoto N., Tomizuka M. Vehicle lateral velocity and yaw
rate control with two independent control inputs. Proc.
ACC1990, San Diego, USA, 1990
Salman M., Zhang Z., Boustany N. Coordinated control of four
wheel braking and rear steering. Proc. ACC1992, Chicago, USA
, 1992
Nagai M., Yamanaka S., Hirano Y. Integrated control law of
active rear wheel steering and direct yaw moment control. Proc.
AVEC96, Aachen, Germany, 1996
Shino M., Nagai M. Independent wheel torque control of
small-scale electric vehicle for handling and stability
improvement. JSAE Review, Vol. 24, No. 4, 2003, pp.449-456
Chu T., Jones R. Integrated four-wheel-steering and direct yaw
moment control via robust eigenstructure assignment techniques.
Proc. AVEC06, Taipei, Taiwan, 2006
Ono E., Takanami K., Iwama N. Vehicle integrated control for
steering and traction systems by mu-synthesis. Automatica, Vol.
30, No. 11, 1994, pp.1639-1647
Cherouat H., Lakehal-Ayat M., Diop S. An integrated braking and
steering control for a cornering vehicle. Proc. AVEC04, Arnhem,
The Netherlands, 2004, pp. 341-346
Boada M., Boada B., Muoz A., Daz V. Integrated control of
front-wheel steering and front braking forces on the basis of fuzzy
logic. Proc. IMechE, Part D: J. Automobile Engineering, Vol.
220, No.3 2006, pp. 253-267
Mokhiamar O., Abe M. Simultaneous optimal distribution of
lateral and longitudinal tire forces for the model following
control. J. Dynamic Systems, Measurement and Control,
Transactions of the ASME, Vol. 126, No.4, 2004, pp. 753-763
Hattori Y., Koibuchi K., Yokoyama T. Force and moment control
with nonlinear optimum distribution for vehicle dynamics. Proc.
AVEC02, Hiroshima, Japan, 2002
Shen X., Yu F. Study on vehicle chassis control integration based
on a main-loop-inner-loop design approach. Proc. IMechE, Part
D: J. Automobile Engineering, Vol. 220, No.11, 2006, pp.
1491-1502
Li D., Shen X., Yu F. Integrated vehicle chassis control with
main/servo-loop structure. Int. J. Automotive Technology, Vol. 7,
No. 7, 2006, pp. 803-812
Shen X., Yu F. Study on vehicle chassis control integration based
on vehicle dynamics and separate loop design approach. Int. J.
Vehicle Autonomous Systems, Vol. 5, No.1-2, 2007, pp. 95-118
Nouillant C., Assadian F., Moreau X. A cooperative control for
car suspension and brake systems. Int. J. Automotive Technology,
Vol. 3, No. 4, 2002, pp. 147-155
Feng J., Yu F., Li J. An investigation on integrated control of
vehicle anti-lock braking system and active suspension.
Transactions of the Chinese Society of Agricultural Machinery,
Vol. 33, No. 2, 2002, pp. 15-19 (in Chinese)
Guo J., Li Y., Li J. Research on integrated control of vehicle
anti-lock braking system and active suspension. China
Mechanical Engineering, Vol. 18, No. 24, 2007, pp.3014-3018 (in
Chinese)
[35] Valasek M., Vaculin O., Kejval J. Global chassis control:
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
integration synergy of brake and suspension control for active

safety. Proc. AVEC04, Arnhem, The Netherlands, 2004, pp.
495-500
Rengaraj C., Adgar A., Cox S., Crolla D. Integration of yaw
control, active suspension and anti-lock brake system to improve
ride and handling. FISITA Congress 2006, Yokohama, Japan,
2006
March C., Shim T. Integrated control of suspension and front
steering to enhance vehicle handling. Proc. IMechE, Part D: J.
Automobile Engineering, Vol. 221, No. 4, 2007, pp. 377-391
Smakman H. Functional integration of active suspension with slip
control for improved lateral vehicle dynamics. Proc. AVEC00,
Ann Arbor, Michigan, USA, 2000
Hac A., Bodie M. Improvements in vehicle handling through
integrated control of chassis systems. Int. J. Vehicle Autonomous
Systems, Vol. 1, No. 1, 2002, pp. 83-110
Kou Y. Integration chassis control (ICC) systems of Mando. SAE
Paper 2004-01-2044, 2004, pp.
Cooper N., Manning W., Crolla D., Levesley M. Integration of
active suspension and active driveline to ensure stability while
improving vehicle dynamics, SAE Paper 2005-01-0414, 2005
Lee A. Coordinated control of steering and anti-roll bars to alter
vehicle rollover tendencies. J. Dynamic Systems, Measurement,
and Control, Transactions of the ASME, Vol. 124, No. 1, 2002,
pp. 127-132
Wang Q., Liu L., Chen W. Integrated control of automotive
electrical power steering system and suspension system based on
random sub-optimal control. China Mechanical Engineering, Vol.
16, No. 8, 2005, pp. 743-747 (in Chinese)
Chen W., Wang Y., Wang Q. Study on multi-variable adaptive
integrated control of automobile electric power steering and
active suspension systems. J. Vibration Engineering, Vol. 18, No.
3, 2005, pp. 360-365 (in Chinese)
ACKNOWLEDGEMENTS
The authors would like to acknowledge the support of
Nature Science Foundation of China (Grant No. 50575141)
and the Opening Fund (Grant No. 20071111) by Key State
Laboratory of Automobile Dynamic Simulation, Jilin
University, www.ascl.jlu.edu.cn .
Jurnal
Teknologi
Full paper
Review of Course Keeping Control System for Unmanned Surface Vehicle

M. N. Azzeria, F. A. Adnanb*, M. Z. Md. Zaina
a
Faculty of Mechanical Engineering, Universiti Teknologi Malaysia, 81310 UTM Johor Bahru, Johor, Malaysia
Technology Center, Transportation Research Alliance, Universiti Teknologi Malaysia, 81310 UTM Johor Bahru, Johor, Malaysia
bMarine
*Corresponding author: faizul@utm.my
Article history
Abstract
Received :25 December 2014

Received in revised form :
25 March 2015
Accepted :15 May 2015
This paper presents a review of research work done on various aspects of control system approaches of
unmanned surface vehicle (USV) in order to improve the course keeping performance. Various methods
have been used to produce a course keeping control system for manoeuvring system of USV. However,
the review reveals that the adaptive backstepping control system is a powerful tool for the design of
controllers for nonlinear systems or transformable to form a tight feedback parameter. It is very suitable
for the automated control system of USV in relative motion that involves the disturbances from waves
and wind. Fuzzy logic control also had been suggested as an alternative approach for complex systems
with uncertain dynamics and those with nonlinearities. This method does not rely on the mathematical
models, but the heuristic approach. Further studies may be conducted to combine the control method
approach mentioned above to develop a real time system with robust control laws to the motions of a
USV in waves, usually at a specific speed, including station keeping or heading in sinusoidal and irregular
waves.
Graphical abstract
Keywords: Unmanned surface vehicle; course keeping; control approach

2015 Penerbit UTM Press. All rights reserved.
1.0 INTRODUCTION
Unmanned surface vehicle (USV) fills an increasingly important
knowledge in the marine engineering study for robotic vehicles.
Research and development in USV is a specialization in the realm
of marine robotics which consists of guidance, navigation, and
control as basic approach for controlling surface vehicles [1].
USVs are also called Autonomous Surface Crafts (ASCs) or
Autonomous Surface Vehicles (or Vessels) (ASVs) [2-6], is a
type of vessel that runs on the surface of water without an
operator on the platform and allows new modes of operations [2].
On the design of the USV, there are mostly unique small
designs with special purpose craft and not standardized for
modularity [6]. This design has limited endurance, payload, and
seakeeping ability. In the current design of USV, most small
vehicles such as boats have been adapted from ships manned
originally designed to accommodate a human occupant [7]. This
design provides moderate speed with long endurance and
improved capabilities of payload and seakeeping to support
special missions.
Thus, USVs can be used for a variety of missions that will be
safer and cheaper than human, such as marine environment
monitoring, hydrology survey, target object searching,
bathymetric mapping, defence, general robotic researches, and
scientific studies. [6, 7]. Some of the missions above need
advanced control system to manoeuvre the USV. Therefore, this
paper is intended to present a literature review of research work
done by many researchers concerning various aspects of control

system of USV technology in an effort to improve the course
keeping performance of its application.
2.0 HISTORY AND DEVELOPMENT

The history of the development of USVs has actually started since
World War II, used for minesweeping purposes and battle damage
assessments [6]. It is only during the last decade that they have
been considered for more advanced operations. However, it is
only in the past few years that USVs have begun to have an
impact in many mission areas, including military purposes and
science marine research studies.
The USVs available in the market today are mainly used in
the military, especially for intelligence, surveillance, and
reconnaissance (ISR) [6, 8]. The unmanned surface vehicles
developed for military purposes such as the SSC San Diego, based
on a sport boat with a jet drive Bombardier SeaDoo Challenger
2000 [9] with robust USV operation in a real world environment,
primarily focus on autonomous navigation, obstacle avoidance,
and path planning. The Israeli Stingray USV [10] and Protector
USV [10] are equipped with electro-optic sensors, radar, Global
Positioning System (GPS), inertial navigation system, and a
stabilized machine gun. Another military USV, already operated
by the British Royal Navy named Shallow Water Influence
Minesweeping System (SWIMS) [10] is used to support the Mine
74:5 (2015) 1120 | www.jurnalteknologi.utm.my | eISSN 21803722 |
12
M. N. Azzeri, F. A. Adnan & M. Z. Md. Zain / Jurnal Teknologi (Sciences & Engineering) 74:5 (2015), 1120
Counter Measures (MCM) operations. SWIMS is basically a boat

that can be converted into Boat Combat Support functions that
need additional equipment. USV has also been developed under
the Officer of Naval Research (ONR) USA named Owl and
Spartan Scout. These USVs are primarily used to conduct
minefield reconnaissance, swallow water monitoring, maritime
interception, and safeguard ports and surrounding areas [11].
For research purpose, USVs have been developed and
demonstrated by some academic institutions, companies, and
government agencies. Most scientific USVs are just experimental
platforms or prototypes and no applications currently exist in the
commercial market [8, 12].
The number of USV or ASC prototypes have been developed
such as the catamarans Autonomous Coastal Exploration System
(ACES) [12] and AutoCat [13] from Massachusetts Institute of
Technology (MIT), which are devoted to the collection of
hydrographic and bathymetric data. Another prototype of ASC
that has been developed by the MIT is Surface Craft for
Oceanographic and Undersea Testing (SCOUT), is a single hull
for oceanographic and undersea testing that supports multiple
ASCs working in cooperative autonomy [14]. The Charlie
catamaran, originally designed by CNR-ISSIA, Genova, Italy, is
used for the sampling of the sea surface micro layer in the
Antarctica [15]. The University of South California has developed
an unmanned airboat for surface water biological study [16, 17]
and has conducted researches on surface obstacle avoidance using
a single beam sonar [18]. A solar-energy-powered ASC designed
by the University of Queensland has integrated various sensors to
do water researches [19]. The DELFIM, developed by the
Institute for Systems and Robotics, Lisboa Portugal, is a surface
platform for underwater communication for automatic marine data
acquisition, and to serve as an acoustic relay between submerged
craft and a support vessel [20]. The ASIMOV ASC [21] project is
initiated by the Commission of the European Communities
regarding to applications of autonomous surface craft which
emphasized ocean observations and data transmission researches.
An Italian catamaran USV, developed by CNR-ISSIA, Genoa
called SESAMO, has been used in the Antarctica in support of
oceanographic researches [22]. A long range of research vessel
ASC Caravela, developed by the Instituto Superior Tcnico (IST)
in Lisbon [6], is used for testing craft in vessel mission control
and radar based obstacle avoidance. The University of Plymouth,
UK has developed autonomous catamaran Springer for sensing,
monitoring, and tracking water pollution [23, 24]. In China, a
water-jet propelled USV also has been developed and it looked
into the feasibility study on different sailing states of the motion
control strategy of the USV [25]. The University Technology of
Malaysia also produced a USV based on modified jet-ski hull
structure for sea patrol and environmental monitoring [26].
Virginia Tech has developed a Ribcraft boat for riverine USV
with robust and accurate abilities for autonomous operation to
execute manoeuvres [27].
Another series of USVs have been developed from the
companies such as USV-2600, by SeaRobotics, have been used in
transportation and bathymetric surveying [28]. The Z-Boat, which
is produced by the Ocean Science company, is a small USV
platform used for hydrographic surveying [28]. The company of
Liquid Robotics produced a small ocean-patrolling USV for ocean
observing [28].
3.0 TECHNICAL DEVELOPMENT

The technical development of USVs can be classified into two
main categories, which are vehicle control technology, and task
oriented design technology [11]. The vehicle control technologies
consist of automatic collision avoidance systems, automatic
navigation systems, and intelligent control systems. The task
oriented is entirely according to the mission requirements to
conduct operations under adverse sea conditions and in areas that
are unsafe for manned vehicles.
The main thing in the technical development of USV is
related to the steering control system. The steering control system
is linked to the course keeping control of USV that can be
controlled remotely or autonomously. The typical steering control
system of the boat using a hydraulic system includes the
component of the helm controller, hydraulic motor pump, cylinder
actuator, servo motor, rudder, and close loop hydraulic line
system, as shown in (Figure 1).
The control of USV is very important and a lot of studies
have been conducted lately on course keeping control system to
improve system performance and speed of action. Therefore,
interest in control system of USV has increased and USV control
systems are now an important area of research [29].
Figure 1 Typical steering control system of boat using hydraulic system

[30]
The challenge brought about by the control system

configuration of the surface vessel that involves motion on three
degrees of freedom that only two actuators are available for
controlling these. The propeller thrust is commonly used for
controlling the surge speed of the surface vessel. This will require
rudder action to control both the sway displacement and the
heading angle. The block diagram in (Figure 2) shows the block
representations of rudder actuator control from the input surge
speed and yaw angle. A reasonable approach to empower the
rudder to simultaneously compensate for the sway displacement
and heading angle is to couple the ship controller with a guidance
system [31-35]. This task is complicated by the ship dynamics,
which are highly nonlinear and involve significant modelling
imprecision. Therefore, effective surface vessel controllers must
be robust for modelling imprecision and capable of adapting to
their unpredictable environmental conditions [31, 36-43].
13
Figure 2 Block representation of rudder actuator control
The dynamic equations of motion for a ship exposed to

waves have evolved from two main directions, which is
manoeuvring theory, and seakeeping theory. In manoeuvring
theory, it is common to assume that the ship is moving at the
restricted calm water. Hence, the ship model is derived for a ship
moving at positive speed U under the zero-frequency assumption
such that added mass and damping can be represented by using
hydrodynamic derivatives. Seakeeping analysis is used in
operability calculations to obtain operability diagrams according
to the adopted criteria. It also refers to the motions of a vessel in
waves usually at a specific speed including station keeping and
heading in a sinusoidal, irregular or random seaway [44].
Nonlinear control theory has also been extensively used in
dynamic equations of motion for a ship exposed to waves of
marine vessels [32, 36, 39, 45-49].
Typically, the most important performance criterion for the course

keeping manoeuvre is minimum course deviation with smallest
control exerted by the steering machine which controls the rudder
angle (). However, it is desirable to track the new set course as
quickly as possible with minimum overshoot [50].
Automatic system for course-keeping or autopilot is
normally based on feedback from a gyrocompass measuring the
heading. Heading rate measurements can be obtained from rate
sensor, gyro, and numerical differentiation of the heading
measurement or a state estimator. This is a common practice in
most control laws utilizing proportional, derivative, and integral
action [50].
The first autopilots implemented on ships were PID-based
autopilots [53]. Later, new types were introduced, including gain
scheduling adaptive autopilots, where autopilot parameters change
with the speed of the vehicle. The first model reference and selftuning adaptive autopilots appeared in the mid-1970s and 1980s,
respectively [54]. During the 1990s, researches were focused on
fuzzy and neuro course-keeping control systems [55].
However, the advanced technique approach using
compensators, such as state feedback linearization techniques [32,
46], output feedback controllers, and back-stepping schemes [36,
39, 47, 48] are model based schemes to enhance the robust system
to modelling uncertainties. However, PID-type autopilots are still
very popular and represent a majority of autopilots in use.
5.0 COURSE KEEPING CONTROL APPROACHES

4.0 PRINCIPLES OF COURSE KEEPING CONTROL
SYSTEM
Hence, ship manoeuvring is treated as a horizontal plane motion,
and only the surge, sway and yaw models are considered. The
main aim of the course-keeping control system is to maintain the
reference course of the vehicle ( = const.) [50, 51]. The
course-keeping control system prevails among control systems
used in guidance of marine vehicles. The vehicles course is
usually measured by gyrocompass. The block diagram in (Figure
3) shows the course keeping control system for automatic
heading. Here, the yaw rate (usually measured by rate gyro) and
surge speed are most often used. Surge speed is important because
the hydrodynamic parameters of the vehicle depend on it [51].
The system is usually designed in such a way that the ship can
move forward with constant speed U, while the sway position is
controlled by correction of rudder angle () using steering system
[52]. The output from the system will then represent the desired
course angle (). The necessity of course-keeping control systems
has become important because of safety and economic reasons
[51].
Figure 3 Course keeping control system for automatic heading block

diagram
Here, various methods have been used to produce a course

keeping control system for USV. Thus, this paper discusses the
methods of control such as control system design using a variety
of methods appropriate to the mission or the use of USV.
Researchers have employed various control approaches to
tackle this problem. A sampling of the research was done for
different control approaches, as shown in (Figure 4) One of the
technologies that have been applied in the various aspects of
course keeping control is to compensate for the error in position
and reduce it as much as possible. Brief review of the idea to
develop a procedure to compensate for the error in position and
how they are employed in course keeping control are given below.
14
obtain using a PID controller, with two fixed thrusters, and is

coupled with the dynamics of the surge and yaw motion. In order
to solve this problem, the optimal Linear-Quadratic (LQ)
controller based on the estimated model was proposed. The
simulation results show that the control objective was achieved
successfully.
Jutong et al., [58] proposed of a Trimaran Unmanned
Surface Vehicle (TUSV) control system and a control scheme for
such a vehicle. The PID controller was designed to control engine
speed, yaw and position. The full system was tested successfully
in the manual operation and obtained useful data, which was
analysed and used in identifying the TUSV model and design
advanced control technologies.
5.2 Optimal Control Methods
Figure 4 Sampling of course keeping control approaches
Optimal control deals with the problem of finding a control law

for a given system, as such that a certain optimality criterion is
achieved. An optimal control is a set of differential equations
describing the paths of the control variables that minimize the cost
function [59]. There are two types of optimal control method that
have been used in the USV control system, which are LinearQuadratic-Gaussian control (LQG), and Model Predictive Control
(MPC).
5.1 Classical Control Methods Based on PID Control
5.2.1 Linear-Quadratic-Gaussian Control (LQG)
A proportional-integral-derivative controller (PID controller) is a

generic control loop feedback mechanism (controller) that is
widely used in industrial control systems. A PID controller
calculates an error value as the difference between a measured
process variable and a desired set point. The controller attempts to
minimize the error by adjusting the process control inputs [50].
The first autopilots implemented on ships were PID-based
autopilots [53].
Kumar et al., [56] reported that a classical PID controller is
the most popular controller due to its simplicity of operation and
low cost. The classic PID controllers are effective for linear
systems, but not suitable for nonlinear and complex systems. As
for nonlinear and complex system, fuzzy logic is used to enhance
the system due to its ability to translate the control action into the
rule base [56].
Caccia et al., [10] demonstrated, in particular, experiments of
a USV through extension at sea trials carried out with the
prototype autonomous catamaran Charlie. The experiments
demonstrate the effectiveness both for the precision and power
consumption of simple PID guidance with Kalman filter and
control laws to perform basic control tasks such as auto-heading,
auto-speed, and straight line, following a USV equipped only with
GPS and compass.
Curcia et al., [14] developed a simple PID guidance
implemented on the SCOUT kayak platform. The navigation
system was further extended by incorporating the distributed
autonomy architecture for sensor adaptive control of USVs in an
autonomous oceanographic sampling scenario.
Park et al., [57] studied dynamic positioning and a waypoint
tracking control experiment to confirm the control performance of
designing a PID controller in the sea-trial. The catamaran USV
has been developed in carrying out experiments to validate the
automatic control performance to keep the USVs position at a
fixed point and to track predefined positions. The optimal control
performance for the speed and heading control was difficult to
Sharma et al., [60] reported that in autonomous catamaran

Springer control design, an LQG controller was selected which
consists of a linear combination of a Linear Quadratic state
feedback Regulator (LQR) and a Kalman filter. It concerns
uncertain linear systems disturbed by additive white Gaussian
noise, having incomplete state information, and undergoing
control subject to quadratic costs.
Naeem et al., [24] investigated the LQG controller in
autonomous catamaran Springer that required a state space model
of the system in the form of specified equation. The LQR and
Kalman filter were developed independently and then combined
to form an LQG controller. In control theory, the LQG control
method is one of the most fundamental optimal control problems
for linear systems. Athans [61] studied that the LQG controller is
also fundamental to the optimal control of perturbed non-linear
systems.
5.2.2 Model Predictive Control (MPC)
Siramdasu and Fahimi [62] introduced the MPC, which calculates
the future control inputs based on the present state variables by
optimizing a cost function. The fact that the cost function
incorporates input constraints, as well as state errors, in
determining the control inputs were exploited. This method can be
applied to all systems with input saturation and suitable for the
trajectory tracking guidance system.
Sharma et al., [63] used the MPC to generate a prediction of
future behaviour of the USV control system. At each time step,
past measurements and inputs were used to estimate the current
state of the system. An optimization problem was solved to
determine an optimal open-loop policy of the present state.
Annamalai et al., [64] studied to improve performance of
traditional autopilots deterioration due to consistent and persistent
changes in the marine environment. One solution to this problem
is to apply MPC techniques in the design of an autopilot to USV.
15
This study also investigated the performance parameters that

compared the MPC and LQG in the simulation analysis. The
MPCs predictive nature results in the USV had a much smoother
trajectory than that was obtained using the traditional LQG
optimal control methods. It shows that MPC can cope with fast
changing dynamics of the plant and can perform well when
integrated with the other subsystem in USV.
5.3 Adaptive Control Methods
Adaptive control is the control method used by a controller which
must adapt to a controlled system with parameters which vary or
are initially uncertain. Adaptive control is concerned with control
law changing themselves as it does not need priori information
about the bounds on these uncertain or time-varying parameters
[65]. There are two types of optimal control methods that have
been used in the USV control system, which are Adaptive
Backstepping, and Gain Scheduling.
5.3.1 Adaptive Control Based On Backstepping Control Design
Adaptive backstepping is a tool for the design of controllers for
nonlinear systems or transformable to form a tight feedback
parameter. Many studies have been conducted to achieve high
performance for ship course keeping control system.
Junfang et al., [66] introduced a robust adaptive control
algorithm for ship course autopilot with parameter uncertainty and
input saturation that applied on USV. The controller was
constructed by considering parameter uncertainty and actuator
saturation constraints using Lyapunov stability theory and
adaptive backstepping technique. Simulation results indicate the
effectiveness and the performance of the stability system to the
effect of input saturation constraints.
Junsheng and Xianku [67] proposed backstepping adaptive
course-keeping controller design for USV autopilot. Therefore,
the proposed controller has no need of a priori knowledge about
ship's system dynamics. Simulation study verifies the efficiency
of the ship course-keeping design and command filter that could
bypass the iterative differential manipulations in conventional
ship course adaptive backstepping controller.
Liao et al., [68] proposed a method of backstepping adaptive
dynamical sliding mode control for the path following control
system of the underactuated surface vessel. The system consists of
the nonlinear ship response model and the Serret-Frenet error
dynamics equations. The control system takes into account of the
modelling errors and external disturbances. Using the Lyapunov
function, it was proven that the proposed controller could render
the path following control system globally asymptotically stable.
Simulation results verify that the controller is robust and adaptive
to the systemic variations or disturbances.
Fossen et al., [35] studied a three degree of freedom; surge,
sway, and yaw, on nonlinear controller for path following of
marine craft using only two controls, using the nonlinear control
theory. Path following is achieved by a geometric assignment
based on a line-of-sight projection algorithm for minimization of
the cross-track error in the path. The control laws in surge and
yaw are derived using backstepping technique. This results in a
dynamic feedback controller where the dynamics of the
uncontrolled sway mode enters the yaw control law. A study
involving an experiment with a model ship is included to
demonstrate the performance of the controller and guidance
systems.
Sonnenburg and Woolsey [27] studied the implementation of two

trajectory tracking control algorithms which is a cascade of
proportional-derivative controllers and a nonlinear controller
obtained through backstepping. This study focused on providing
riverine USV with the ability to robustly and accurately execute
manoeuvres. Experimental results indicate that the backstepping
controller is much more effective at tracking trajectories with high
variable speed and course angle.
In control theory, backstepping is a technique for designing
stabilizing controls for nonlinear dynamical systems [69, 70]
using the Lyapunov function method to stabilize the subsystem
controllers that progressively stabilize each outer subsystem [71].
5.3.2 Adaptive Control Based on Gain Scheduling
Gain scheduling is a method used to find auxiliary variables that
correlate well with the changes in the process dynamics
implemented in computer-controlled systems. By changing the
parameter variations, which are the functions of auxiliary,
variables feedback gains are adjusted [72].
Murali et al., [72] reported that the main problem in
designing the gain scheduling controller is to find suitable
scheduling variables. This is normally done based on the physics
of the system. When scheduling variables have been deformed,
the controller parameters are calculated at a number of operating
conditions using some suitable design approach. The controller is
tuned for each operating condition. The stability and performance
of the system are evaluated at different operating conditions.
The experiments of two types of controllers above, Adaptive
Backsteepping, and Gain Scheduling, were performed on boat
model, conducted by Murali et al., [72]. The effectiveness of the
two types of controller is validated for reaching the desired
heading angle that satisfies the steering dynamics and the first
order of the Nomotos model. The result indicated that the desired
angle reaches faster in case of adaptive controller. The adaptive
controller has some notable advantages over the gain scheduling
controller.
5.4 Intelligent Control
Intelligent control is a control technique that uses various
Artificial Intelligence (AI) computing approaches like genetic
algorithms, neural networks, and fuzzy logic [73].
5.4.1 Intelligent Control Based on Genetic Algorithms
Malecki and Zak [74] introduced the genetic algorithms for
designing a fuzzy trajectory control for ships at low speed. For
track-keeping precise control, the waypoint line of sight scheme is
incorporated and three independent fuzzy controllers are used to
generate command signals. In general, the genetic algorithm
techniques manipulate sets of individuals by using genetic
operators in order to propose better ones. The individuals in a
population are represented by chromosomes [75]. An advantage
of this control system is its flexibility with regard to the change of
dynamic properties of the special ship. From the simulation
results presented, the proposed approach provides the autopilot
being robust and having good performance both without and in
the presence of the sea current disturbances [74].
16
5.4.2 Intelligent Control Based on Neural Networks

Zhouhua et al., [76] proposed the adaptive dynamic surface
control for the formation of autonomous surface vehicles by
employing neural network and dynamic surface control technique.
The advantages of the proposed formation controller are that:
first, the proposed method only uses the measurements of line of
sight range and angle by local sensors, no other information about
the leader is required for control implementation; second, the
developed neural formation controller is able to capture the
vehicle dynamics without exact information of coriolis and
centripetal force, hydrodynamic damping, and disturbances from
the environment. Comparative analysis with a model-based
approach is given to demonstrate the effectiveness of the proposed
method. Simulation results demonstrate the effectiveness of the
formation controller and the learning ability of neural networks.
Furthermore, Peng et al., [77] extended the study from
Zhouhua et al., to form a control of autonomous surface vehicles
with uncertain leader dynamics and uncertain local dynamics,
where backstepping based neural networks control technique is
employed to stabilize formations. However, the controller is
complicated for the sake of the needs to calculate numerical
derivatives of virtual control signals. Hence, there is a need to
develop an adaptive formation controller using the neural
networks based dynamic approach to derive the relative dynamics
between the leader and follower from the kinematics.
using robust control methods tend to be able to cope with small

differences between the true system and the nominal model used
for design. The characteristic of the robust system is the norm
with the Input constraints, bounded control action, and closed
loop system properties [79].
On the other hand, Jerzy [80] introduced the design of
robust, nonlinear control system of the ship course angle, in a
model following control (MFC) structure based on an inputoutput linearization. This controller reacts to the difference
between the output signal of real ship and its model. The
nonlinear plant, linearized by the input-output method, has good
properties (short time control, and acceptable control signals).
Therefore, the system proposes the use of MFC structure, which is
able to compensate for differences of nonlinear characteristics of
the process and model and can replace the complex and expensive
adaptive control system.
Shr and Jeng [81] investigated the robust nonlinear ship
course keeping control under the influence of high wind and large
wave disturbances. To design the robust nonlinear ship course
keeping controller employs two feedback loops, which are the H
approximate input/output linearization in the inner-loop, and
synthesis in the outer-loop, to address tracking, regulation, and
robustness issues. Simulation results indicate that the performance
of the proposed robust nonlinear controller design is better than
the linear robust controller in handling high wind and large wave
disturbances.
5.4.3 Intelligent Control Based on Fuzzy Logic
5.6 Sliding Mode Control
Fuzzy logic control has been suggested as an alternative approach

for complex systems with uncertain dynamics and those with
nonlinearities. This method occupies the boundary line between
artificial intelligence and control engineering, and it can be
considered as an obvious solution which is confirmed by
engineering practice [56]. An unconventional control design such
as fuzzy control design methods do not rely on the mathematical
models [50].
Moreover, Nassim and Nabil [49] introduced a self-tuning
fuzzy sliding mode controller for marine surface vessel. It is
designed to control the surge speed and the heading angle of an
under-actuated marine surface vessel. The actuation signals were
restricted to the propeller thrust, and the rudder control torque.
The controller was designed based on a reduced order model,
which only considers the surge and yaw motions of the vessel.
The simulation results illustrate good tracking characteristics of
the current guidance and control system in spite of considerable
modelling imprecision and environmental disturbances. The
results also demonstrate that the proposed guidance scheme
enables the marine vessel to converge to its desired trajectory
faster.
Junsheng and Xianku [78] proposed ship course-keeping
adaptive tracking fuzzy control scheme of the nonlinear system in
strict-feedback form. The control objective is to force the ships
course to track the output of the specified reference model. The
fuzzy system is employed to approximate the unknown dynamics.
The proposed algorithm can guarantee the boundedness of all the
signals in the closed-loop system.
In control theory, sliding mode control is a nonlinear control

method that alters the dynamics of a nonlinear system through the
application of a discontinuous control signal that forces the
system to slide along a cross-section of the system's normal
behaviour. Hence, sliding mode control is a variable structure
control method. The multiple control structures are designed so
that trajectories always move toward an adjacent region with a
different control structure, and so the ultimate trajectory will not
exist entirely within one control structure. Instead, it will slide
along the boundaries of the control structures [82].
Wei et al., [83] proposed a new nonlinear sliding mode based
on formation control scheme for underactuated surface vessels.
Since the sway axis is not directly actuated, the vessels were
underactuated. The formation model is obtained based on the
leader-following approach. The controller was designed using the
Lyapunovs direct method and sliding mode control technique.
The first-order surface in terms of the surge motion tracking
errors and the second-order surface in terms of sway motion
tracking errors were introduced. The effectiveness of the designed
sliding mode formation controller was validated by numerical
simulations.
Ashrafiuon et al., [84] studied a sliding-mode control law on
experimentally implemented trajectory tracking of underactuated
autonomous surface vessels. The control law is developed by
introducing a first-order sliding surface in terms of surge tracking
errors and a second-order surface in terms of lateral motion
tracking errors. The experimental vessel is a small boat with two
propellers in an indoor pool. The control law was developed using
a first-order surface in terms of the surge tracking errors, and a
second-order surface in terms of the sway tracking errors. The
vessels absolute position and orientation were measured using a
camera. The motor input voltages were estimated from the
controller propeller forces and transmitted to the motors using
5.5 Robust Control

Robust methods aim to achieve robust performance and stability
in the presence of small modelling errors. Controllers designed
17
wireless transmitters and receivers. Several straight-line and

circular experiments were successfully performed.
Piotr [85] investigated a course controlling a USV following
three control methods, which are PID, slide, and fuzzy control.
PID, slide, and fuzzy controllers were tested for three different
changes of course from 0 to 30, 90 and 180. Moreover, to
counteract sea current, two methods were presented and compared
by means of numerical research. The comparison results from the
numerical research show that the slide controller achieved more
control quality indexes than the others.
5.7 Summary and Future Work
Hence, conclusions can be derived from the list of citations in
Table 1 on some of the researches done on course keeping control
system for USV. Most of the control approaches were not a real
time experimental implementation of nonlinear dynamic
algorithms. Therefore, no direct attention has been given on the
methods with which parameters of the nonlinear dynamic model
can be identified in the control law with uncertain parameters and
environmental disturbances.
In addition, the marine operations are characterized by timevarying disturbances and widely changing sea conditions. Thus,
the controller design for such nonlinear system is still a
challenging problem. Some attempts have been made to solve the
problem based on the adaptive control. However, the review also
reveals that adaptive backstepping is a powerful tool for the
design of controllers for nonlinear systems or transformable to
form tight feedback parameters. This type of control method
requires much knowledge in mathematical algorithms, involving
the Lyapunov function method, and backstepping technique.
These methods are suitable for the automated control system of
USV in relative motion that involves a disturbance from waves
and wind. Fuzzy logic control also has been suggested as an
alternative approach for complex systems with uncertain
dynamics and those with nonlinearities. The nonlinear and
complex system used fuzzy logic to enhance the system due to its
ability to translate the control action into the rule base. This type
of method does not rely on mathematical models but the heuristic
approach.
Therefore, future research should consider the dynamic

positioning problem to achieve station keeping mode for a USV to
have only two independent control inputs; surge force, and yaw
moment. From the previous researches in Table 1, typically, the
USV is equipped with two aft thrusters located at a distance from
the centre line to give both surge force and yaw moment, but has
no side thruster to give sway force control. To control both
positions, together with the orientation of three degrees of
freedom and having only two independent control inputs
available, the system will experience an underactuated control
problem. The underactuated USV control problem is inherently
nonlinear [86]. Therefore, a nonlinear ship model must be
considered. Moreover, the underactuated ship belongs to a class of
systems that cannot be asymptotically stabilized by a feedback
control law [86]. This implies that the dynamic positioning
problem cannot be solved using a linear proportional integral
derivative (PID)-controller or any other linear time-invariant
feedback control law, and the problem is not solvable using
classic nonlinear control theory like feedback linearization [86].
Thus, station keeping is achieved by applying the necessary
thrust in opposition to any measured external disturbance. For
systems that operate in the ocean, care has to be taken to ensure
that the ship is filtered out with high frequency wave disturbances
which use the acceleration feedback in dynamic positioning. Such
underactuated dynamic positioning describes a feedback control
law that keeps a USV in station in an asymptotic manner using
accurate visual position feedback.
For control strategies of the USV, one of the ways to stay at
the desired location is to arrive at the location in a direction that is
parallel to the disturbance and surging forward or reversing. This
control strategy needs the USV to align with the direction of
current and wave when it is close to the desired target. The
designed controller will try to turn back towards the location if it
drifts away from the target location. The controller will control
the two thrusters with the ability to turn in circles. The movement
of the turn-centre is much smaller than the length of the boat, and
hence for simplicity, it will assume that it turns about the centre of
the desired target.
18
Table 1 Researches on course keeping control system approaches for USV
6.0 CONCLUSION
In this study, the literature related to course keeping control
system for USV has been reviewed. Simple classifications of
different course keeping control approaches have been defined.
The controllers were based on a two dimensional planar model of
the system that included surge, sway, and yaw dynamics. The
review reveals the significant progress made during the last two
decades in the development of USV and the course keeping
control system. However, the review also reveals that adaptive
backstepping is a powerful tool for the design of controllers for
nonlinear systems or transformable to form tight feedback
parameters. This method is suitable for the automated control
system to USV in relative motion that involves a disturbance from
waves and wind. Fuzzy logic control also has been suggested as
an alternative approach for complex systems with uncertain
dynamics and those with nonlinearities. This type of method does
not rely on mathematical models but the heuristic approach.
Furthermore, the review also reveals a weakness in regard to

real time experimental implementation of these control
approaches for USV. The development of real time control laws
that can be implemented to USV with uncertain parameters and
environmental disturbances particularly refers to the motions of a
vessel in waves usually at a specific speed, including station
keeping, heading in a sinusoidal, irregular or random seaway.
Thus, future work should focus on particular problem related
to this study. Further studies may be conducted to combine the
control method approach to be applied in optimal control theory
to find the dynamic positioning problem to achieve station
keeping mode. This course keeping control system can play the
role as an autopilot controller for the future of USV as a maritime
monitoring boat or environmental monitoring boat, and other
capabilities, such as disaster monitoring boat that carries out the
desired command without any involvement of manpower for
control.
19
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
Furfaro, T. C. 2012. A Modular Guidance, Navigation and Control

System for Unmanned Surface Vehicles. Florida Atlantic University,
Master Thesis, Florida, USA.
Manley, J. E. 2008. Unmanned Surface Vehicles, 15 Years of
Development. In: Proceeding Oceans 2008 MTS/IEEE Quebec
Conference and Exhibition, Quebec City, Canada. 14.
Caccia, M. 2006. Autonomous Surface Craft: Prototypes and Basic
Research Issues. In: Control and Automation, 2006. MED '06. 14th
Mediterranean Conference, Ancona. 16.
Blank, J., Bishop, B. E. 2008. In-Situ Modeling of a High-Speed
Autonomous Surface Vessel. In: System Theory, 2008. 40th Southeastern
Symposium System Theory, New Orleans. 347351.
Higinbotham, J. R., Kitchener, P. G., Moisan, J. R. 2006. Development
of a New Long Duration Solar Powered Autonomous Surface Vehicle.
In: OCEANS 2006, Boston, USA. 16.
Bertram, V. 2008. Unmanned Surface VehiclesA Survey. In:
Skibsteknisk Selskab, Copenhagen, Denmark. 114.
Jianhua, W., Wei, G., Jianxin, Z., Jubiao, Z. 2009. An Unmanned Surface
Vehicle for Multi-mission Applications. In: Electronic Computer
Technology, 2009 International Conference Macau, China. 358361.
Breivik, M., Hovstein, V. E., Fossen, T. I. 2008. Straight-Line Target
Tracking for Unmanned Surface Vehicles. J. Modeling, Identification
and Control. 29(4): 131149.
Larson, J., Bruch, M., Ebken, J. 2006. Autonomous Navigation and
Obstacle Avoidance for Unmanned Surface Vehicles. In: Proceedings of
the SPIE, Unmanned Systems Technology VIII. 6230: 722.
Caccia, M., Bibuli, M., Bono, R., Bruzzone, G. 2008. Basic Navigation,
Guidance and Control of an Unmanned Surface Vehicle. J. Autonomous
Robots. 25(4): 349365.
Yang, W. R., Chen, C. Y., Hsu, C. M., Tseng, C. J., Yang, W. C. 2011.
Multifunctional Inshore Survey Platform with Unmanned Surface
Vehicles. International Journal of Automation and Smart Technology.
1(2): 1925.
Manley, J. E. 1997. Development of The Autonomous Surface Craft
"ACES". In: OCEANS '97. MTS/IEEE Conference Proceedings, Halifax,
Canada. 827832.
Manley, J. E., Marsh, A., Cornforth, W., Wiseman, C. 2000. Evolution of
the Autonomous Surface Craft AutoCat. In: OCEANS 2000 MTS/IEEE
Conference and Exhibition, Providence, USA. 403408.
Curcio, J., Leonard, J., Patrikalakis, A. 2005. SCOUT-A Low Cost
Autonomous Surface Platform for Research in Cooperative Autonomy.
In: OCEANS, 2005. Proceedings of MTS/IEEE. 1: 725729.
Caccia, M. 2006. Autonomous Surface Craft: Prototypes and Basic
Research Issues. In: 14th Mediterranean Conference on Control and
Automation, 2006. MED '06, Ancona, Italy. 16.
Dhariwal, A., Sukhatme, G. 2007. Experiments in Robotic Boat
Localization. In: IEEE/RSJ International Conference on Intelligent
Robots and Systems, 2007. IROS 2007. San Diego, USA. 17021708.
Bin, Z., Sukhatme, G. 2007. Adaptive Sampling for Estimating a Scalar
Field using a Robotic Boat and a Sensor Network. In: IEEE International
Conference on Robotics and Automation, 2007 Roma, Italy. 36733680.
Heidarsson, H. K., Sukhatme, G. 2011. Obstacle Detection and
Avoidance for an Autonomous Surface Vehicle Using a Profiling Sonar.
In: Robotics and Automation (ICRA), 2011 IEEE International
Conference on, Shanghai, China. 731736.
Grinham, A., Udy, J., Dunbabin, M. 2009. An Autonomous Surface
Vehicle for Water Quality Monitoring. In: Australasian Conference on
Robotics and Automation (ACRA), Sydney, Australia. 16.
Alves, J., Oliveira, P., Oliveira, R., Pascoal, A., Rufino, M., Sebastiao,
L., Silvestre, C. 2006. Vehicle and Mission Control of the DELFIM
Autonomous Surface Craft. In: 14th Mediterranean Conference on
Control and Automation, 2006. MED '06, Ancona, Italy. 16.
Pascoal, A., Oliveira, P., Silvestre, C., Sebastiao, L., Rufino, M., Barroso,
V. et al. 2000. Robotic Ocean Vehicles for Marine Science Applications:
The European ASIMOV Project. In: OCEANS 2000 MTS/IEEE
Conference and Exhibition, Providence, USA. 409415.
Caccia, M., Bono, R., Bruzzone, G., Bruzzone, G., Spirandelli, E.,
Veruggio, G., Stortini, A. M., Capodaglio, G. 2005. Sampling Sea
Surfaces with SESAMO: An Autonomous Craft for The Study of Sea-Air
Interactions. J. Robotics & Automation Magazine, IEEE. 12(3): 95105.
Naeem, W., Xu, T., Sutton, R., Chudley, J. 2006. Design of an
Unmanned Surface Vehicle for Environmental Monitoring. In: World
Maritime Technology Conference, London, United Kingdom, London,
U.K. 16.
Naeem, W., Xu, T., Sutton, R., Tiano, A. 2008. The Design of a
Navigation, Guidance and Control System for an Unmanned Surface
Vehicle for Environmental Monitoring. Proceedings of the Institution of
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
Mechanical Engineers, Part M: Journal of Engineering for the Maritime

Environment 2008. 222(2): 6779.
Gongxing, W., Hanbin, S., Jin, Z., Lei, W. 2009. The Basic Motion
Control Strategy for The Water-Jet-Propelled USV. In: International
Conference on Mechatronics and Automation, 2009. ICMA 2009,
Changchun, China. 611616.
Yaakob, O., Mohamed, Z., Hanafiah, M., Suprayogi, D., Ghani, M. A., F.
A. Adnan, Mukti, M., Din, J.v 2012. Development of Unmanned Surface
Vehicle (USV) for Sea Patrol and Environmental Monitoring. In:
International Conference on Marine Technology Kuala Terengganu,
Malaysia, Kuala Terengganu, Malaysia. 16.
Sonnenburg, C. R., Woolsey, C. A. 2013. Modeling, Identification, and
Control of an Unmanned Surface Vehicle. J. Field Robotics. 30(3): 371
398.
Wang, M. 2012. The Development of and Experiments on an Unmanned
Surface Vehicle Platform for Water Sampling and Monitoring. Hong
Kong University of Science and Technology, Master Thesis, Hong Kong.
Yulei, L., Lei, W., Jiayuan Z. 2010. Full State-Feedback Stabilization of
an Underactuated Unmanned Surface Vehicle. In: Advanced Computer
Control (ICACC), 2nd International Conference 2010. 4: 7074.
Inboard Power Steering. Retrived on
Dec 20, 2013 from
http://www.seastarsolutions.com/products/hydraulic-new/inboard-powersteering.
Do, K. D., Pan, J., Jiang, Z. P. 2003. Robust Adaptive Control of
Underactuated Ships on a Linear Course with Comfort. J. Ocean
Engineering. 30(17): 22012225.
Moreira, L., Fossen, T. I., Guedes Soares, C. 2007. Path Following
Control System for a Tanker Ship Model. J. Ocean Engineering. 34(14
15): 20742085.
Healey, A. J., Marco, D. B. 1992. Slow Speed Flight Control of
Autonomous Underwater Vehicles: Experimental Results with The NPS
AUV II. In: Proceedings of the 2nd International Offshore and Polar
Engineering Conference (ISOPE), San Francisco, California. 523532.
Zhong-Ping, J., Nijmeijer, H. 1999. A Recursive Technique for Tracking
Control of Nonholonomic Systems in Chained Form. J. Automatic
Control, IEEE. 44(2): 265279.
Fossen, T. I., Breivik, M., Skjetne, R. 2003. Line-of-Sight Path
Following of Underactuated Marine Craft. In: Proceedings of the Sixth
IFAC Conference on Maneuvering and Control of Marine Crafts
(MCMC2003), Girona, Spain. 244249.
Pettersen, K. Y., Nijmeijer, H. 2001. Underactuated Ship Tracking
Control: Theory and Experiments. Int. J. Control. 74(14): 14351446.
Godhavn, J. M., Fossen, T. I. 1998. Nonlinear and Adaptive
Backstepping Designs for Tracking Control of Ships. Int. J. Adapt.
Control Signal Process. 12(Special Issue on Marine Systems): 649670.
Lauvdal, T., Fossen, T. I. 1998. Rudder Roll Stabilization of Ships
Subject to Input Rate Saturations Using a Gain Scheduled Control Law.
In: Proceedings of the IFAC Conference on Control Applications in
Marine Systems (CAMS98), Fukuoka, Japan. 2730.
Fossen, T. I., Strand, J. P. 1999. A Tutorial on Nonlinear Backstepping:
Applications to Ship Control. J. Model, Identification and Control. 20(2):
83135.
imen, T., Banks, S. P. 2004. Nonlinear Optimal Tracking Control with
Application to Super-Tankers for Autopilot Design. J. Automatica.
40(11): 18451863.
Li, Z., Sun, J., Oh, S. 2009. Design, Analysis and Experimental
Validation of a Robust Nonlinear Path Following Controller for Marine
Surface Vessels. J. Automatica. 45(7): 16491658.
Yang, Y., Bo, J. 2004. Robust Adaptive Fuzzy Control (RAFC) for Ship
Steering with Uncertain Nonlinear Systems. In: Intelligent Control and
Automation, 2004. WCICA 2004. Fifth World Congress on. 3: 2514
2518.
Yang, Y., Zhou, C., Ren, J. 2003. Model Reference Adaptive Robust
Fuzzy Control for Ship Steering Autopilot with Uncertain Nonlinear
Systems. Applied Soft Computing. 3(4): 305316.
Fossen, T. I. 2005. A Nonlinear Unified State-Space Model for Ship
Maneuvering and Control in A Seaway. J. Bifurcation and Chaos. 15(9):
27172746.
Pivano, L., Johansen, T. A., Smogeli, O. N., Fossen, T. I. 2007.
Nonlinear Thrust Controller for Marine Propellers in Four-Quadrant
Operations. In: American Control Conference, 2007. ACC '07. 900905.
Berge, S. P., Ohtsu, K., Fossen, T. I. 1999. Nonlinear Control of Ships
Minimizing the Position Tracking Errors. J. Model Identification and
Control. 20(3): 177187.
Fossen, T. I., Grovlen, A. 1998. Nonlinear Output Feedback Control of
Dynamically Positioned Ships Using Vectorial Observer Backstepping. J.
Control Systems Technology, IEEE. 6(1): 121128.
20
[48] Strand, J. P., Ezal, K., Fossen, T. I. 1998. Nonlinear Control of Ships: A
Locally Optimal Design. In: Preprints of The IFAC NOLCOS98,
Enschede, Netherlands. 732738.
[49] Nassim, K., Nabil, G. C. 2013. A Self-Tuning Guidance and Control
System for Marine Surface Vessels. J. Nonlinear Dynamics. 73(12):
897906.
[50] Fossen, T. I. 1994. Guidance and Control of Ocean Vehicles. John Wiley
and Sons, Chichester, England.
[51] Zoran, Vc, Bruno, Bc. 2001. Guidance and Control Systems for Marine
Vehicles. in The Ocean Engineering Handbook, F. El-Hawary, Ed. 1st
ed. Boca Raton CRC Press. 2527.
[52] Holzher, T., Schultze, R. 1996. Operating Experience with a HighPrecision Track Controller for Commercial Ships. Control Engineering
Practice. 4(3): 343350.
[53] Minorsky, N. 1992. Directional Stability of Automatically Steered
Bodies. J. American Society for Naval Engineers. 34(2): 280309.
[54] Van Amerongen, J. 1982. Adaptive Steering of Ships-A Model Reference
Approach to Improved Maneuvering and Economical Course Keeping,
Delft University of Technology, Ph.D. Thesis, Delft, Netherlands.
[55] Hearn, G. E., Zhang, Y., Sen, P. 1997. Alternative Designs of Neural
Network Based Autopilots: A Comparative Study. In: Proc. of 4th IFAC
Conference on Maneuvering and Control of Marine CraftMCMC 97,
New York, USA.
[56] Kumar, V., Nakra, B.C., Mittal, A. P. 2011. A Review on Classical and
Fuzzy PID Controllers. Int. J. Intelligent Control and Systems. 16(3):
170181.
[57] Park, J-H, Shim, H-W, Jun, B-H, Kim, S-M, Lee, P-M, Lim Y-K. 2010.
A Model Estimation and Multi-Variable Control of an Unmanned
Surface Vehicle with Two Fixed Thrusters. In: OCEANS 2010 IEEE
Sydney. 15.
[58] Juntong, Q., Yan, P., He, W., Jianda, H. 2007. Design and Implement of
a Trimaran Unmanned Surface Vehicle System. In: International
Conference on Information Acquisition, 2007. ICIA '07, Seogwipo,
South Korea. 361365.
[59] Donald, E. K. 1998. Optimal Control Theory: An Introduction. PrenticeHall, New Jersey, U.S.A
[60] Sharma, S. K., Naeem, W., Sutton, R. 2012. An Autopilot Based on a
Local Control Network Design for an Unmanned Surface Vehicle. J.
Navigation. 65(2): 281301.
[61] Athans, M. 1971. The Role and Use of The Stochastic Linear-QuadraticGaussian Problem in Control System Design. J. Automatic Control,
IEEE. 16(6): 529552.
[62] Siramdasu, Y., Fahimi, F. 2012. Incorporating Input Saturation for
Underactuated Surface Vessel Trajectory Tracking Control. In: American
Control Conference (ACC), 2012, Montreal, Canada. 62036208.
[63] Sharma, S. K., Sutton, R., Motwani, A., Annamalai, A. 2013. Non-Linear
Control Algorithms for an Unmanned Surface Vehicle. Proceedings of
the Institution of Mechanical Engineers, Part M. Journal of Engineering
for the Maritime Environment 2008. 227(4): 110.
[64] Annamalai, A., Motwani, A., Sutton, R., Yang, C., Sharma, S.,
Culverhouse, P. 2013. Integrated Navigation and Control System for an
Uninhabited Surface Vehicle Based on Interval Kalman Filtering and
Model Predictive Control. In: IET Conference on Control and
Automation 2013. 16.
[65] Astrom, K. J., Wittenmark, B. 1995. Adaptive Control. Addison-Wesley,
Reading, Massachusetts.
[66] Junfang, L., Tieshan, L., Zhongzhou, F., Renxiang, B., Qiang, L.,
Jiangqiang, H. 2011. Robust Adaptive Backstepping Design for CourseKeeping Control of Ship with Parameter Uncertainty and Input
Saturation. In: International Conference of Soft Computing and Pattern
Recognition (SoCPaR), Dalian, China. 6367.
[67] Junsheng, R., Xianku, Z. 2013. Fuzzy-Approximator-Based Adaptive

Controller Design for Ship Course-Keeping Steering in Strict-Feedback
Forms. J. Applied Sciences, Engineering and Technology. 6(16): 2907
2913.
[68] Liao, Y-l, Wan, L., Zhuang, J-y. 2011. Backstepping Dynamical Sliding
Mode Control Method for the Path Following of the Underactuated
Surface Vessel. J. Procedia Engineering. 15(2011): 256263.
[69] Kokotovic, P. V. 1992. The Joy of Feedback: Nonlinear and Adaptive. J.
Control Systems, IEEE. 12(3): 717.
[70] Lozano, R., Brogliato, B. 1992. Adaptive Control of Robot Manipulators
with Flexible Joints. J. Automatic Control, IEEE. 37(2): 174181.
[71] Khalil, H. K. 2002. Nonlinear Systems. 3rd Edition. Prentice Hall, Upper
Saddle River, New Jersey
[72] Murali, K., Sundarasiva, R., Ramachandra, R. 2011. Theoretical and
Experimental Investigation of Gain Scheduling and Adaptive Autopilots
for A Model Boat. International Journal of Engineering Science and
Technology (IJEST). 3(2): 902911.
[73] Spooner, J. T., Maggiore, M., Ordonez, R., Passino, K. M. 2002. Stable
Adaptive Control and Estimation for Nonlinear Systems: Neural and
Fuzzy Approximation Techniques. John Wiley and Sons, New York.
[74] Malecki, J., Zak. B. 2012. Design a Fuzzy Trajectory Control for Ship at
Low Speed. In: 12th WSEAS International Conference on Signal
Processing, Computational Geometry and Artificial Vision (ISCGAV
'12), Istanbul, Turkey. 119123.
[75] Kacprzyk, J. 1996. Multistage Fuzzy Control: A Model-Based Approach
to Fuzzy Control and Decision Making. John Wiley and Sons, New York.
[76] Zhouhua, P., Dan, W., Zhiyong, C., Xiaojing, H., Weiyao, L. 2013.
Adaptive Dynamic Surface Control for Formations of Autonomous
Surface Vehicles With Uncertain Dynamics. J. Control Systems
Technology, IEEE. 21(2): 513520.
[77] Peng, Z., Wang, D., Hu, X. 2011. Robust Adaptive Formation Control of
Underactuated Autonomous Surface Vehicles with Uncertain Dynamics.
J, Control Theory & Applications, IET. 5(12): 13781387.
[78] Junsheng, R., Xianku, Z. 2013. Ship Course-Keeping Adaptive Fuzzy
Controller Design Using Command Filtering with Minimal
Parametrization. In: 25th Chinese Control and Decision Conference 2013
(CCDC) Guiyang, China. 243247.
[79] Mackenroth, U. 2004. Robust Control Systems: Theory and Case Studies.
Springer-Verlag Berlin Heidelberg, New York, U.S.A
[80] Jerzy, B. 2012. Design of Robust, Nonlinear Control System of the Ship
Course Angle, in a Model Following Control (MFC) Structure Based on
an Input-Output Linearization. J. Scientific Maritime University of
Szczecin. 30(102): 2529.
[81] Shr, S. H., Jeng, Y. J. 2011. Robust Nonlinear Ship Course-Keeping
Control Under The Influence of High Wind and Large Wave
Disturbances. In: Control Conference (ASCC), 2011 8th Asian. 393398.
[82] Utkin, V. I. 1993. Sliding Mode Control Design Principles and
Applications to Electric Drives. IEEE Transactions on Industrial
Electronics. 40(1): 2336.
[83] Wei, M., Chen, G., Yang, L. 2012. Nonlinear Sliding Mode Formation
Control for Underactuated Surface Vessels. In: 10th World Congress on
Intelligent Control and Automation (WCICA), 2012 Beijing, China.
16551660.
[84] Ashrafiuon, H., Muske, K. R., McNinch, L. C., Soltan, R. A. 2008.
Sliding-Mode Tracking Control of Surface Vessels. IEEE Transactions
on Industrial Electronics. 55(11): 40044012.
[85] Piotr, S. 2013. Course Control of Unmanned Surface Vehicle. Solid State
Phenomena. 196(2013): 117123.
[86] Pettersen, K. Y., Fossen, T. I. 2000. Underactuated Dynamic Positioning
of a Ship-Experimental Results. IEEE Transactions on Control Systems
Technology. 8(5): 856863.

49
24
2013
12
Vol.49 No.24
Dec.
2013
JOURNAL OF MECHANICAL ENGINEERING
DOI
10.3901/JME.2013.24.108
NVH
1, 2 1
201804
2. 153-8505 )
(1.
NVH
(Noise, vibration and harshness, NVH)
NVH

U46
Review of Dynamic Research for NVH Problems

Related to Automotive Driveline
WU Guangqiang
1, 2
LUAN Wenbo
(1. Automotive School, Tongji University, Shanghai 201804;

2. Institute of Industrial Science, University of Tokyo, Tokyo 153-8505, Japan)
AbstractDifficulties, characteristics and mechanism of several typical NVH problems related to automotive driveline, advances in
present overseas and domestic researches are concluded in this paper. The prime research contents and routes of vehicle driveline
NVH problems are also introduced. The discussions are unfolded from two aspects of simulation and experiment. Aiming at dynamic
research of NVH problems related to automotive driveline, the characteristics of four mechanical models which are systematic
dynamics model, multi-body dynamics model, structural dynamics model, systematic and structural dynamics hybrid model are
analyzed in detail, coordinates and reference systems used in modeling and the main simulation quantitative and qualitative research
methods are also reviewed in simulation. Meanwhile, research contents, signal processing methods, subjective and objective
assessment are presented briefly in experiment research aspect.
Key wordsDriveline of automotive Vibration Noise Harshness Dynamic model
(Noisevibration and
harshnessNVH)NVH
NVH
(51175379)20130715 20131128
NVH NVH
[1]
2013 12
NVH
109
NVH
NVH
NVH
NVH
NVH
NVH
NVH NVH
(1)
(2)
(3) (
)

NVH
NVH
NVH
NVH
NVH 2016 000 Hz

(
) 1
(
1
NVH
1 (FF)
1

NVH

[2]
[3]
110
[5]
(Automated mechanical transmission
AMT) 3
[6](Torque converterTC)
15
[9]
[10]
4
(Dual mass flywheelDMF)

NVH

[4]
[7]
[8]
[11]
[12]
[13]
[14]
49 24
[15]
[16]
(50500 Hz)
[19]
()(
)
[20]
(Total value analysisTVA)

(Failure mode and effects analysis
FMEA)
NVH
NVH
NVH
[17]
[18]
NVH
NVH
( 2)
2013 12
NVH
NVH
(2)
(3)
(4)
(1)
NVH
111
NVH
3 NVH
NVH
()()
()
()--
()
NVH
3.1
NVH
NVH
(
)()

[21]
112
NVH
3.2
3.3
3.4
49 24
NVH
()
(
)
3.5
NVH
NVH
NVH
(1) ()
()
-
adams
-
-
4.1
[22]
2013 12
NVH

NVH
(2)
(3)
()

(
)
-
[23]
[24]
[25]
4.2
[25]
NVH
NVH
113
(
)
NVH
NVH
NVH
NVH
(Root mean square, RMS)
Gabor
46
[26]
[14]
114
NVH
[27]
[1]
[2]
10
NVH
(1) NVH
NVH
(2) NVH
(3)
NVH
NVH

[3]
49 24
[4]
[5]
[6]
, , .
[J]. , 2000, 21(3)22-26.
JIANG Guoping WANG Guolin, CHEN buda. Review
on research of vibration of vehicle powertrain[J]. Journal

of Jiangsu University of Science and Technology, 2000,
21(3)22-26.
CROWTHER A, ZHANG N, LIU D K, et al. Analysis
and simulation of clutch engagement judder and stick-slip
in automotive powertrain systems[J]. Journal of
Automobile Engineering, 2004, 218(12)1427-1446.
ALBERS A, HERBST D. Chatter-causes and
solutions[C]// The 6th LuK Symposium, 1998. Germany
Luk GmbH & Co, 199823-45.
HIROYUKI K, KATSUYUKI H. A prediction method of
vehicle vibration caused by the drive torque fluctuation at
takeoff[C]//Society of Automotive Engineers. SAE
Technical Papers. Warrendale, PA, USA SAE
International, 2007 2007-01-3499.
NAUS G J L, BEENAKKERS M A, HUISMAN R G M,
et al. Robust control of a clutch system to prevent
judder-induced driveline oscillations[J]. Vehicle System
Dynamics, 2010, 48(11)1379-1394.
TAKAHIRO R, KENICHIRO M, TAKASHI N, et al. A
study on shudder in automatic transmission lock-up clutch
systems and its countermeasures[C]//Society of
Automotive Engineers. SAE Technical Papers.
2013 12
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
115
NVH
USA SAE International, 2011
nonlinear vibration of gear pairs[J]. Journal of Sound and
Vibration, 2012, 331(15)3628-3648.
Warrendale, PA,
2011-01-1509.
WANG M Y, MANOJ R, ZHAO W. Gear rattle
modelling and analysis for automotive manual
transmissions[J]. Proceedings of the Institution of
Mechanical Engineers, Part DJournal of Automobile
Engineering, 2001, 215(2)241-258.
, , , .
[J]. , 1997, 16(3)391-396.
WANG Jianjun, LU Mingwan, ZHANG Xiong, et al. The
theory of rattle dynamics for automotive gear shift
system[J]. Mechanical Science Technology, 1997, 16(3)
391-396.
XIANG Dong, XIE Ning, SHEN Gang, et al. The
influence of rotation speed on gear rattle under
overdrive[C]// 3rd International Conference on Advanced
Engineering Materials and Technology, AEMT 2013.
Advanced Materials Research. SwitzerlandTrans. Tech.
Publications Ltd., 2013753-755, 960-967.
MEHMET B. Torsional vibration model based
optimization of gearbox geometric design parameters to
reduce rattle noise in an automotive transmission[J].
Mechanism and Machine Theory, 2010, 451583-1598.
, , . DMF-CS
[J]. , 2009, 20(15)
1787-1790.
LI Wei, LONG Yan, SHI Wenku. Analysis of isolation of
the torsional vibration of DMF-CS with centrifugal
pendulum-type absorber[J]. China Mechanical Engineering,
2009, 20(15)1787-1790.
, .
[J]. , 2011, 31(6)76-80.
SHI Wenku, MIN Delei. Study on vibration damping
characteristics of triple mass flywheel torsional damper[J].
Noise and Vibration Control, 2011, 31(6)76-80.
CURTIS S, PEARS J, PALMER D. An analytical method
to reduce gear whine noise, including validation with test
data[C]//Society of Automotive Engineers. SAE Technical
Papers. Warrendale, PA, USA SAE International, 2005
2005-01-1819.
.
[J]. , 2013, 32(11)95-99.
LUAN Wenbo, WU Guangqiang, XU Weiqing. Gearbox
whine noise analysis based on order tracking[J]. Journal
of Vibration and Shock, 2013, 32(11)95-99.
ERITENEL T, PARKER R G. Three-dimensional
[16] FIETKAU P, BERTSCHE B. Influence of tribological and

geometrical parameters on lubrication conditions and
noise of gear transmission[J]. Mechanism and Machine
Theory, 2013, 69303-320.
[17] , , . [J].
, 2009, 26(2)124-128.
MENG Xianjie, WANG Huan, WU Guangqiang. Review
of research on brake groan[J]. Journal of Highway and
Transportation Research and Development, 2009, 26(2)
124-128.
[18] FUADI Z, ADACHI K, IKEDA H. Experimental model
for creep groan analysis[J]. Lubrication Science, 2009,
21(1)27-40.
[19] FUADI Z, MAEGAWA S, NAKANO K, et al. Map of
low-frequency stick-slip of a creep groan[J]. Journal of
Engineering Tribology, 2010, 224(12)1235-1246.
[20] KARABAY S, BAYNAL K, IGDELI C. Detecting groan
sources in drum brakes of commercial vehicles by
TVA-FMEA A case study[J]. Journal of Mechanical
Engineering, 2013, 59(6)375-386.
[21] . [M].
, 1990.
FU Zhifang. Vibration modal analysis and parameter
identification[M]. BeijingChina Machine Press, 1990.
[22] , . (MATLAB )[M].
2008.
MA Changfeng, LIN Weichuan. Modern numerical
calculation method[M]. BeijingScience Press, 2008.
[23] . [M]. , 2011.
ZHANG Yimin. Mechanical vibration[M]. Beijing
Tsinghua University Press, 2011.
[24] , . [M].
2001.
LIU Yanzhu, CHEN Liqun. Nonlinear vibration[M].
BeijingHigher Education Press, 2001.
[25] , , . [M].
1998.
LIU Yanzhu, CHEN Wenliang, CHEN Liqun. Mechanics
of vibration[M]. BeijingHigher Education Press, 1998.
[26] . [M].
, 2004.
HU Guangshu. Modern signal processing tutorial[M].
BeijingTsinghua University Press, 2004.
[27] , , . - [M].
116
, 2006.
PANG Jian, CHEN Gang, HE Hua. Automotive noise and

vibration-principle and application[M]. BeijingBeijing
Institute of Technology Press, 2006.
1965
Vehicle System DynamicsInternational Journal of Vehicle

Design
220 80 SCI/Ei
(
)Ground Vehicle Dynamics
49 24
16
4 21 2
3 SAE
1 2 4
2 1
2
E-mailwuguangqiang@tongji.edu.cn
()1985
7 4
Ei 1
E-mailflashlwb_168@163.com
September, 2009
Int J Agric & Biol Eng
Open Access at http://www.ijabe.org
Vol. 2 No.3
Review of research on agricultural vehicle autonomous guidance

Ming Li, Kenji Imou, Katsuhiro Wakabayashi, Shinya Yokoyama
(Graduate School of Agricultural and Life Sciences, the University of Tokyo, Japan)
Abstract: A brief review of research in agricultural vehicle guidance technologies is presented.
The authors propose the
conceptual framework of an agricultural vehicle autonomous guidance system, and then analyze its device characteristics.
This paper introduces navigation sensors, computational methods, navigation planners and steering controllers. Sensors
include global positioning systems (GPS), machine vision, dead-reckoning sensors, laser-based sensors, inertial sensors
and geomagnetic direction sensors. Computational methods for sensor information are used to extract features and fuse
data. Planners generate movement information to supply control algorithms.
into changes in position and direction.
proceeded to commercialization.
Actuators transform guidance information
A number of prototype guidance systems have been developed but have not yet
GPS and machine vision fused together or one fused with another auxiliary
technology is becoming the trend development for agricultural vehicle guidance systems.
Application of new popular
robotic technologies will augment the realization of agricultural vehicle automation in the future.
Keywords: agricultural vehicle, guidance, machine vision, GPS
DOI: 10.3965/j.issn.1934-6344.2009.03.001-016
Citation: Ming Li, Kenji Imou, Katsuhiro Wakabayashi, Shinya Yokoyama. Review of research on agricultural vehicle
autonomous guidance.
Int J Agric & Biol Eng, 2009; 2(3): 116.
the 1920s[2,3].
Introduction
The concept of fully autonomous
agricultural vehicles is far from new; examples of early
Over history, agriculture has evolved from a manual

occupation to a highly industrialized business, utilizing a
wide variety of tools and machines[1].
driverless tractor prototypes using leader cable guidance

systems date back to the 1950s and 1960s [4].
Researchers are
In the 1980s, the potential for combining computers
now looking towards the realization of autonomous
with image sensors provided opportunities for machine
agricultural vehicles.
vision based guidance systems.
The first stage of development,
During the mid-1980s,
automatic vehicle guidance, has been studied for many
researchers at Michigan State University and Texas A&M
years, with a number of innovations explored as early as
University were exploring machine vision guidance.

Also during that decade, a program for robotic harvesting
Received date: 2009-07-20

Accepted date:2009-08-30
Biography: Ming Li, PhD candidate in Graduate School of
Agricultural and Life Sciences, the University of Tokyo. His
research interests include agricultural robotics, navigation and
computer vision.
Email:
liming@bme.en.a.u-tokyo.ac.jp.
Kenji Imou, PhD, Associate Professor in Graduate School of
Agricultural and Life Sciences, the University of Tokyo. Phone:
81-03-5841-5359, Fax: 81-03-5841-8173. Email: aimou@mail.ecc.
u-tokyo.ac.jp. Katsuhiro Wakabayashi, MS student in Graduate
School of Agricultural and Life Sciences, the University of Tokyo.
Email: aa096212@mail.ecc.u-tokyo.ac.jp. Shinya Yokoyama,
Professor in Graduate School of Agricultural and Life Sciences, the
University of Tokyo. Email: syokoyama@bme.en.a. u-tokyo.ac.jp
of oranges was successfully performed at the University

of Florida[5].
In 1997, agricultural automation had
become a major issue along with the advocacy of

precision agriculture.
The
automated
vehicles
productivity,
agricultural
increased
potential
include
application
enhanced operation safety.
benefits
of
increased
accuracy,
and
Additionally, the rapid
advancement in electronics, computers, and computing

technologies has inspired renewed interest in the
development of vehicle guidance systems.
Various
September, 2009
Vol. 2 No.3
guidance technologies, including mechanical guidance,
navigation system for agricultural vehicles is now
optical guidance, radio navigation, and ultrasonic
regarded as an important advance in precision agriculture
[6,7]
guidance, have been investigated
and a promising alternative to the dwindling farming
Table 1 summarizes examples of research systems

that have been developed around the world. Autonomous
labor force, in addition to satisfying the quest for higher

production efficiency and safer operation [6, 8].
Table 1 Examples of guidance systems developed around the world

Institute (Country)
Sensor
Machine or test device
Performance results
Literature
University of Illinois, USA
Machine vision, GPS,

GDS
Case 8920 MFD and 2WD

Tractors
Vision guidance at 16 km/h on row crops
Zhang[9,10]
Benson[11,12]
Stanford University, USA
GPS
John Deere 7800 Tractor
1 accuracy in heading, line tracking accuracy with

2.5 cm deviation
OConnor[13]
University of Florida, USA
GPS, laser radar
Tractor
Average error of 2.8 cm using machine vision guidance

Subramanian[14]
and average error of 2.5 cm using radar guidance
University of Halmstad, Sweden
Machine vision,
Mechanical sensor, GPS
Tractor with row cultivator
Standard deviation of
position of 2.7 and 2.3 cm
strand
Bygholm Research Center,

Denmark
Machine vision
Tractor
Accuracy of less than 12 mm
Sgaard[17]
University of Tokyo, Japan
FOG, Ultrasonic
Doppler sensor
Tractor (Mitsubishi Co.)
Lateral displacement from the reference line was less

than 10 cm at speeds of 0.7 to 1.8 m/s on a straight line
Imou[18]
National Agriculture Research

Center, Japan
RTK GPS, FOG
PH-6, Iseki Co., Ehime

transplanter
Less than 12 cm, yaw angle offset of about 5.5 cm at

2.52 km/h
Nagasaka
BRAIN, Japan
Machine vision and

laser range sensor
Tractor
Error about 5 cm at the speed of 0.4 m/s
Yukumoto[20]
Hokkaido University, Japan
GDS, laser scanner
Tractor
Average error less than 1 cm
Noguchi [21,22]
National Centre for Engineering

in Agriculture, Australia
Machine vision
Tractor
Accuracy of 2 cm
Billingsley[23]
Research on autonomous agricultural vehicles has
[15,16]
[19]
set of circumstances from that encountered by a
become very popular, and the robotics industry has
laboratory mobile robot.
developed a wide range of remarkable robots.
In the
simplified by the absence of clutter typically present in
near future, farmers will be using affordable, dependable
the indoor environment; however, a number of additional
autonomous vehicles for agricultural application.
complications are raised.
In one respect, operation is
For example, the operating
Section 2 includes an analysis of the device
areas are large; ground surfaces may be uneven;
characteristics of agricultural vehicle guidance systems.
depending on the operation, and wheel slippage may be
A brief overall review of the past 20 years of global
far from negligible. Cultivation may interfere with
research in agricultural vehicle guidance technologies is
underground cables, colors may change with plant growth,
presented in terms of a framework for agricultural vehicle
and soil quality may vary. Environmental conditions (rain,
autonomous guidance systems, as shown in Figure 1.
fog, dust, etc.) may affect sensor function; moreover, a
The key elements are navigation sensors, computational
low-cost system is required.
methods, navigation planners and steering controllers.
These disadvantages make it more difficult to realize
The final section addresses some of the barriers to
agricultural automation.
Companies are unwilling to
development and discusses the potential for new
invest in commercialization because it is not seen as a
development.
worthwhile money-making venture, and farmers are not

financially able to participate.
Features of agricultural vehicle devices

The agricultural environment offers a very different
Other major reasons
include the need to improve the technology and decrease

the cost [24].
September, 2009
Vol. 2 No.3
understanding of the geometric relationship between the

image sensor, the vehicle and the field-of-view that the
sensor uses for guidance information.
Figure 2 shows
one example. Researchers have explored the use of vision

sensors for detecting a guidance directrix on row crops,
soil tillage, and the edges along harvested crops.
Various methodologies of image processing have been
investigated for extracting the guidance information.
The processed images provide output signals that can be
used to provide steering signals for the vehicle.
Figure 2 Row detection via segmented binary image [25]

Figure 1 Framework of agricultural vehicle autonomous guidance
system
One of the most commonly used machine vision

methods is for detecting a guidance directrix on row
factors,
crops, soil tillage, and the edges along harvested crops.
agricultural farm fields have several advantages for
Benson[12] developed a guidance combine harvester based
developing autonomous guidance systems.
on the lateral position of the crop cut edge.
Compared
with
these
complicating
For example,
[26]
Marchant
the working areas generally do not change; landmarks can
and Brivot
be easily set up around the corners of a field and be taken
in real time (10 Hz) and noted that their technique was
as a stationary environment.
The crops are always the
tolerant to outliers (i.e., weeds) only when their number
same plants at the same places and can be easily
was reasonably small compared to the number of true
distinguished.
data points.
Therefore, even though there are more
used the Hough transform for row tracking
Marchant [27] reported an overall RMS error
disadvantages than advantages for realizing agricultural
of 20 mm in the lateral position at a travel speed of
vehicle autonomous guidance, there are enough research
0.7 m/s using this technique to guide an agricultural
achievements to promote its development.
vehicle in a transplanted cauliflower field.

The threshold method has been applied in many
Navigation sensors
3.1 Machine vision
vision applications to separate objects of interest from

imagery.
For reliably extracting crop row features from
Machine vision sensors measure the relative position
field images, the major challenge of the threshold method
and heading using the image sensor mounted on the
is the difficulty in determining an adequate effective
vehicle.
threshold value under varying ambient light conditions or
There are several aspects of machine vision
based sensing.
Different types of sensor modalities can
changing crop growth stages.
The effectiveness of
be selected to measure the guidance information.
distinguishing crops from weeds is another challenge in
Positioning of the sensor on the vehicle requires an
determining a pathway using the obtained field images.
September, 2009
Vol. 2 No.3
Research has been reported on attempts to improve the
obtained a standard error in hoe position of 13 mm at
reliability of crop feature extraction and pathway
travel speeds up to 6 km/h. Hague and Tillett [28] used the
determination
systems.
analysis of the periodic near-infrared intensity function in
exploited a method using a bandpass
a lateral path across five wheat rows in a plane view of
filter to attenuate the grey level of weeds and shadows in
the field rather than a traditional row segmentation
for
Hague and Tillett
[28]
vision-based
[29]
field images. Pinto et al.
guidance
attempted to apply the
principal component analysis method to extract crop row

[7]
features from field images. Sgaard and Olsen
also
method to obtain row guidance information.
They
obtained a root-mean-square (RMS) position error of 15.6

mm at a travel speed of 5.8 km/h.
developed a machine vision guidance method that did not
For more complete crop or field information, some
require a plant segmentation step, replacing it with a less
researchers used a stereovision system to provide a
intensive computation of the center of gravity for row
three-dimensional (3D) field image by combining two
segments in the image and weighted linear regression to
monocular field images taken simultaneously from a
determine the position and orientation of the rows.
binocular camera.
Han et al.
[25]
developed a row segmentation algorithm
based on k-means clustering to segment crop rows.

information was then used to steer a tractor.
This
Such 3D images are reconstructed
based on the different-disparity monocular images to

Kise et al.[32]
decrease the ambient light influence.
The guided
developed a stereovision-based agricultural machinery
tractor was able to perform field cultivation in both
crop-row tracking navigation system. The RMS error of
straight and curved rows.
Okamoto et al.
[30]
developed
lateral deviation was 35 cm following both straight and
an automatic guidance system for a weeding cultivator.
curved rows at speeds up to 3.0 m/s.
A color CCD camera acquired the crop row images,
required some weed-free areas to provide sufficient
which were then processed by computer and used to
information for detecting the navigation points.
determine the offset between the machine and the target

crop row.
The method
strand and Baerveldt[16] developed a machine vision

guidance system that achieved good performance in
Other techniques and systems have been investigated
detecting plants in near-infrared images acquired under
for machine vision guidance, and many of them have
non-uniform natural illumination by performing grayscale
improved the robustness and dependability of machine
opening on the raw near-infrared image and subtracting it
vision.
Yukumoto et al.
[20]
developed a tillage robot
with vision and laser range sensor.
They used laser
sensor to improve the robustness.
Billingsley and
[23]
Schoenfisch
reported a vision guidance system that is
from the original prior to segmentation.

based upon the
Their method,
Hough transform, used multiple
rectangular regions (one for each row viewed) with the

rectangle width adjusted for crop size.
The information
relatively insensitive to additional visual noise from
from multiple rows was fused together to obtain a
weeds.
common estimate of the row position.
They used linear regression in each of three crop
The accuracy of
row segments and a cost function analogous to the
position estimation was less than 1.2 cm with a standard
moment of the best-fit line to detect lines fitted to outliers
error depending on plant size.
(i.e., noise and weeds) as a means of identifying row
the system had sufficient accuracy and speed to control
guidance information.
the cultivator and mobile robot in a closed-loop fashion
They showed that their system is
capable of maintaining an accuracy of 2 cm.

[31]
Field tests showed that
with a standard deviation of position of 2.7 and 2.3 cm,
developed a machine vision
respectively, with incomplete row structures due to
guidance system for cereal crops, using the midpoints of
missing plants combined with high weed pressure (up to
15 rows extracted from a single view of three adjacent
200 weeds/m2).
Tillett and Hague
They tested the
Kaizu and Imou [33] developed a dual-spectral camera
system in a single barley field with light to moderate
system, shown as Figure 3, for paddy rice seedling row
weed pressure under uniform natural lighting and
detection.
crop rows (five midpoints per row).
The system used a pair of low-cost
September, 2009
Vol. 2 No.3
monochrome cameras with different spectral filters. It
guidance systems is that they can be easily programmed
matched a near-infrared image and a red image and it
to follow curved rows[40].
worked in the strong reflections on the water surface

under cloudy conditions from morning to dusk.
There appear to be three limitations to using GPS for

vehicle guidance.
The first is that GPS guidance
systems cannot be used in microwave-shielded areas.

Also, GPS cannot promise consistent positioning
accuracy in the range of centimeters for a variety of field
conditions (e.g., presence of buildings, trees or steeply
rolling terrain, and interruption in satellite or differential
correction signals).
The second limitation is the
inherent time delay (data latency) required for signal

processing to determine locations that might present
control system challenges at higher field speeds. The
third is the high cost for agricultural application (although
Figure 3 Schematic diagram of dual-spectral camera system[33]
there is a consistent trend of cost reduction with

widespread use).
3.2
Global positioning system
technology
However, with the anticipated
developments,
these
limitations
will
Since the early 1990s, Global Positioning System
undoubtedly be overcome, thereby making GPS a choice
(GPS) receivers have been widely used as global
candidate for incorporation into vehicle guidance
guidance sensors
[34-37]
. GPS-based guidance technology
systems.
can be used for many field operations such as sowing,
Stoll and Kutzbach[41] studied the use of the RTK GPS
tilling, planting, cultivating, weeding and harvesting[38,39].
as the only positioning sensor for the automatic steering
GPS-based navigation systems are the only navigation
system of self-propelled forage harvesters.
technologies that have become commercially available
that the standard deviation of steering was less than
for
100 mm under all conditions.
farm
vehicles.
Many tractor manufacturing
They found
Standard deviation of
companies now offer the Real-Time Kinematic (RTK)
lateral offset (error) along straight-line paths ranged from
GPS based auto steering system as an option on their
25 to 69 mm depending upon the travel speed.
The position information from the RTK GPS
Kise et al.[42,43] studied the use of an RTK GPS
can be used for both guidance and other applications such
guidance system for control of a tractor as an autonomous
as seed mapping, traffic control, and tillage control.
vehicle traveling along a curved path.
GPS guidance systems provide an absolute guidance
following a sinusoidal path with a 2.5-m amplitude and
system in contrast to the relative guidance provided by
30-m wavelength at 6.5 km/h showed a 6-cm RMS error
machine vision, which requires that the crop be planted
with a 13-cm maximum error. To compensate for GPS
using a GPS-guided planting system or the crop rows
positioning error associated with machinery attitude,
mapped using some type of geo-referenced mapping
researchers at Hokkaido University integrated an inertial
technique.
measurement unit (IMU) with an RTK GPS to provide
tractors.
GPS guidance systems also require that a
Test results for
GPS base station be located within approximately 10 km
more accurate navigation information.
of the RTK GPS guided tractor or agricultural robot.
navigation system could guide agricultural machinery
However, since GPS systems do not depend upon the
performing all field operations, including planting,
visual appearance of the crop, they are not adversely
cultivating and spraying, at a travel speed of up to 3 m/s,
affected by weed density, shadows, missing plants or
with a tracking error of less than 5 cm on both straight
other conditions that degrade the performance of machine
and curved paths.
vision guidance systems.
Another advantage of GPS
This integrated
Ehsani et al.[44] evaluated the dynamic accuracy of
September, 2009
Vol. 2 No.3
several low-cost GPS receivers with the position
which increase proportionally with the distance traveled
information from an RTK GPS as reference.
by the robot.
They
Despite these limitations, researchers use
found that these receivers had an average absolute
odometry as an important part of robot navigation
cross-track error of around 1 m when traveling in a
systems[45,46].
GPS cannot be effectively used for
Doppler sensors use the principle based on the
positioning in citrus applications since the vehicle
Doppler shift in frequency observed when radiated
frequently moves under the tree canopy, which blocks the
energy reflects off a surface that is moving relative to the
satellite signals to the GPS receiver. Moreover, a system
emitter.
using GPS for guidance requires that a predetermined
errors arising from wheel slippage, tread wear, and/or
path be given for the vehicle to follow. Consequently,
improper tire inflation.
significant time must be spent in mapping its path.
autonomous tractor using an ultrasonic Doppler speed
straight line.
[19]
Nagasaka et al.
used an RTK GPS for positioning,
This type of sensor can decrease some of the

Imou et al.[18] developed an
sensor and gyroscope.
The results showed that the
and fiber optic gyroscope (FOG) sensors to maintain
maximum lateral displacement from the reference line
vehicle inclination, for an automated six-row rice
was less than 10 cm at a speed of 4 steps from 0.7 to
transplanter (Figure 4).
1.8 m/s on 50-m straight driving tests.
Root-mean-square deviation
from the desired straight path after correcting for the yaw
Imou et al. [47,48] developed a new ultrasonic Doppler
angle offset was approximately 55 mm at a speed of
sensor to achieve high accuracy when measuring the
0.7 m/s.
speeds of both forward and reverse motions including
The maximum deviation from the desired path
low-speed motions.
was less than 12 cm.
3.4
Laser-based sensors
Laser-based sensors have a relatively longer range and
higher resolution.
The guidance systems need three or
more reflectors (landmarks) around the work field. The

time at which the laser beam is detected is communicated
to the guidance system, which uses triangulation to define
the location of the vehicle. The system is insensitive to
environmental conditions, e.g., strong light change for
machine vision and microwave shadowing for GPS,
which will make the system inoperable.
Figure 4 Automated rice transplanter
[19]
However,
laser-based sensor systems have two drawbacks.
They
do not work well if the position is changed for any of the

3.3
Dead-reckoning sensors
artificial landmarks.
If natural landmarks are used in the
Dead-reckoning sensors are inexpensive, reliable
navigation process, map updating is necessary in order to
sensors for short-distance mobile robots, using a simple
register the landmarks in the map building operation. The
mathematical procedure for determining the present
second problem is noisy laser measurements when the
location of a vehicle by advancing a previous position
vehicle is traveling on uneven ground.
through a known course and velocity information over a

given length of time.
The simplest form of dead
reckoning is referred to as odometry.
Holmqvist [49] used a laser-optic navigation system for

a vehicle moving at a speed of 2 m/s.
With an average
However,
distance to the reflectors of 50 m, the absolute position
odometry is the integration of incremental motion
error will typically be about 5 cm in each of the X, Y and
information over time, which inevitably leads to the
Z directions. Ahamed et al.
unbounded
developing a positioning method using reflectors for
accumulation
of
errors.
Specifically,
orientation errors will cause large lateral position errors,
infield road navigation.
[50]
used laser radar for
They tested differently shaped
September, 2009
reflectors to determine the accuracy in positioning.

Junya et al.
[51]
used a single-laser distance sensor for
Vol. 2 No.3
the use of a heading gyro (e.g., Imou et al. [18]; Barawid et

al. [52]; Ishida et al. [57]).
Inertial sensors are mostly used in combination with
vehicle navigation experiments, in which the vehicle
Zhang and Reid [9] presented an
repeated stop-and-go driving, stopping every 1 m for a
GPS or machine vision.
distance of 20 m.
on-field navigation system with a vision sensor, FOG and
The calculated RMS localization error
in stopping was about 6 mm in the traveling direction and
RTK GPS.
about 12 mm in the transverse direction.
based agricultural navigation system was capable of
Because the tree canopy frequently blocks the satellite

microwaves to the GPS receiver, laser-based sensors are
[52]
The results indicated that the multiple sensor
guiding a tractor between crop rows and showed that the

inertial sensor was a good assistant function.
developed
Noguchi et al. [21] developed an agricultural navigation
an automatic guidance system for navigating between tree
system consisting of an RTK GPS and an inertial
rows.
measurement unit. Experiments conducted in a soybean
widely applied in orchards. Barawid et al.
Their research used a 56-kW agricultural tractor,
2D laser scanner, RTK GPS and FOG.
The results
field for tilling, planting, cultivating and spraying
showed an accuracy of 11 cm lateral error and 1.5
demonstrated that the accuracy of the vehicle surpassed
[14]
heading error. Subramanian
developed an autonomous
that of skilled farmer operation.
guidance system for citrus grove navigation based on
guided vehicle was less than 5 cm.
machine vision and laser radar.
3.6
An average error of 2.8
cm using machine vision guidance and 2.5 cm using
The lateral error of the
Geomagnetic direction sensor (GDS)

A
geomagnetic
direction
sensor
(GDS)
is
radar guidance was observed during vehicle testing on a
magnetometer that senses the earths magnetic field. It
curved path at a speed of 3.1 m/s.
can be used as a heading sensor similar to an electronic
Tofael
[53]
developed a complex autonomous tractor
system with a laser rangefinder, RTK GPS and gyroscope.

The results of field experiments using the laser
compass[6].
The GDS is generally used to supplement
other sensors.
Noguchi et al.[58] used a GDS to provide heading
[11]
rangefinder showed a lateral error of less than 2 cm and a
information to a tillage robot. Benson et al.
heading error of less than 1. The accuracy was very
GPS with GDS for vehicle guidance along straight
high.
directional lines.
3.5
One limitation of GDS sensors is the
influence of external electromagnetic interference from
Inertial sensors
Inertial sensors take measurements of the internal
state of the vehicle.
used
A major advantage of inertial
the outside environment, such as from a nearby set of

high-tension electrical wires or the vehicle heater/air
sensors is that they are packaged and sealed from the
conditioner fan.
environment, which makes them potentially robust under
sources, they were able to combine GDS with a
harsh environmental conditions.
medium-accuracy GPS system (20 cm) and track a
types
of inertial sensors
gyroscopes.
The most common
are accelerometers
Accelerometers
measure
relative to an inertial reference frame.
and
straight line with an average error of less than 1 cm.
acceleration
The maximum overshoot for a 3-m step response was
This includes
gravitational and rotational acceleration as well as linear

acceleration.
Gyroscopes measure the rate of rotation
independent of the coordinate frame.
However, by controlling these error
They can also
12%, compared to 50% for GPS alone.

The feasibility of correlating GDS with sensor
applications for agricultural guidance systems has been
researched.
Harper and Mckerrow[59] used a frequency-
provide 3D position information and have the potential to
modulated ultrasonic sensor to detect plants, setting up a
detect wheel slippage.
plant database with a return signal containing information
Unfortunately, these types of
sensors are prone to positional drift
[54]
Inertial sensors have been used in a number of vehicle

[19,55,56]
applications
. The most common application is in
about the geometric structure of the plants to improve

navigation.
Yekutieli and Pegna [60] used a sensing arm
to detect plants in the path for guidance in a vineyard.
September, 2009
Vol. 2 No.3
However, using an arm would require that citrus groves
determination for crop row detection.
be even with continuous canopy.
dealt with crop row detection for autonomous tractor
There are also
concerns about damaging the tree branches.
Ultrasonic
The research also
guidance.
sensors are used for guidance in greenhouses, but they
strand et al.[15] modeled a plant row using a
require that the target be perpendicular to the sensor for
rectangular box instead of a line. The width of the box
the ultrasonic waves to be reflected back properly
is equal to the average width of the plants and the length
[14]
). Dead reckoning is also widely
of the box is unlimited as it fills the whole image.
used in combination with other sensors for autonomous
The rectangular box can be described by a set of parallel
(Subramanian et al.
vehicles (e.g., Morimoto et al.
[61]
).
adjacent lines, which appear in the image as a set of lines

that intersect at a virtual point outside the image, as
Computational methods
shown in Figure 5.
A computational method is mainly to detect image

features by image processing or deal with sensor data
fusion successfully for providing with basic information
for agricultural vehicle autonomous guidance system.
Therefore, the method choice and improvement is very
important.
4.1
Hough transform
The Hough transform technique can be used to isolate
the features of a particular shape within an image.
The
transform was originally concerned with the identification
Figure 5 Rectangular box substitutes for a line [15]
of lines in the image, but later it was extended to

identifying the position of arbitrary shapes, most
commonly circles or ellipses.
4.2
Kalman filter
The Hough transform as it
The Kalman filter[64] provides a sound theoretical
is universally used today was developed in 1972 by
framework for multi-sensor data fusion. The approach
Richard Duda and Peter Hart, who called it a "generalized
depends upon tracking the position of the vehicle or the
Hough transform" after the related 1962 patent of Paul
state of the system at all times.
V.C. Hough.
often applied in GPS receivers to provide position
The main advantage of using a Hough
Kalman filter models are
transform is that it is quite robust even if a group of
estimates from raw GPS signals.
points varies to some extent, and seeking a straight line is
system that has the potential for significant acceleration,
still possible.
it is necessary to integrate GPS with an Inertial
The disadvantage is that in order to plot
In a highly dynamic
curves (i.e., sinusoids) for every observation point in
Navigation
Cartesian image space to r in the polar Hough
Literature on the integration of INS and/or other sensors
parameter space, the load of computational complexity is
with GPS is abundant[19,38,39,65-68].
large.
are
systems can improve the positioning accuracy, and more
a number of publications on deriving guidance signals
importantly, can provide reliable short-term positioning
from
information if the GPS signal is lost.
As most crops are cultivated in rows, there

plant
structures
[15,26,27,52,56,62,63]
transform
using
the
Hough
[32]
tractor automated guidance
used a stereovision-based
elevation
functions
map
using
Kalman
filters.
These integrated
measurement data and effectively removed the DGPS

noise
and reduced the
root-mean-squared (RMS)
The algorithm
positioning error. The maximum cross-tracking error was
of stereo-image processing,
reduced from 9.83 to 2.76 m and the root-mean-squared
agricultural machinery guidance system.

of
(INS)
Han et al.[25] applied Kalman filtering to raw DGPS
A stereovision-based crop-row detection method for
consists
System
creation
and
navigation
point
error was reduced from 0.58 to 0.56 m.
September, 2009
Hague and Tillett[28] provided a method in which
Vol. 2 No.3
error of 1.0 cm from 30 images.
The corn field results
image processing was combined with a bandpass filter
showed an average RMS offset error of 2.4 cm from 15
and extended Kalman filter.
images.
The method does not rely
upon segmentation of the plant background to reduce the

brightness or color influence.
Results are shown in
Figure 6.
Figure 7 Middle line is the guiding row line [7]
Navigation planners
Navigation planner plays an important role for
agricultural
vehicle
autonomous
control,
which
transforms the position deviation (heading, position or

state) of the vehicle or device into the steering angle.
Besides including tracking methods, the navigation
planning must consider the sensor information and
Fig.6
Row location
vehicle motion to guidance in the desirable course.
[28]
5.1
Navigation planning uses four methods: position
A new sigma-point Kalman filter was proposed and

used to improve the Kalman filter
[69]
. Zhang et al.
[70]
Tracking methods
tracking, line tracking, map tracking and obstacle
compared both filters through simulation and found that
avoidance.
the sigma-point Kalman filter was more numerically
some nominal trajectory or directrix line. The method
robust and computationally efficient.
usually uses local information including crop rows, swath
4.3
edges, and tilled/untilled boundaries.
Other methods
[7]
Most guidance system operations follow
However, if the
proposed a method based on
tracking signal weakens or vanishes, the operation fails.
machine vision for detection and localization of crop
Map tracking is often used in GPS systems, but it is a
rows distinguished by using the generalized Hough
labor- and time-intensive method.
transformation method (as shown in Figure 7).
5.2
Sgaard and Olsen
The
Vehicle motion models
method divided the grayscale image into horizontal strips
1) Dead reckoning
and computed the center of gravity, by vector, as a
Dead reckoning is reliable for short-distance traveling
substitute for the segmentation step to reduce the
(two positions) on a smooth concrete road. Since motion
computational burden on the image processing.
information is integrated in order to obtain the position of
[25]
used three methods to obtain a guidance
the vehicle, there is a risk of error accumulation leading
directrix, which applied a k-means clustering algorithm
to positional drift if the sensor produces even a slight bias.
for row segmentation, a moment algorithm for row
On agricultural vehicles, dead-reckoning sensors can be
detection, and a cost function for guidance line selection.
as simple as wheel encoders, which measure the rotation
The soybean field results showed an average RMS offset
of the vehicle or equipment wheels.
Han et al.
Freeland et al.[71]
10
September, 2009
Vol. 2 No.3
experimented with a low-cost electronic compass used
machine vision, GPS and GDS) since an individual
together with wheel encoders to provide dead-reckoning
sensing technology cannot satisfy vehicle automation
position information.
navigation operation for all models and all methods of
Dead reckoning is widely used in
combination with other sensors for autonomous vehicles.

[19]
Nagasaka et al.
[72]
and Kodagoda et al.

[73]
encoders. Garaca-Prez et al.
used rotary
used odometers and a
use in different environments.
The appropriate sensor
will function at the appropriate field status during

operation.
Nevertheless, even under a given field
proximity capacitive sensor.
operation, the availability of data from multiple sensors
2) Kinematic model
provides the opportunity for better data integration to
Kinematic models are very simple and have been used
provide superior results compared to those using an
by researchers to describe the lateral error relative to a
individual sensor. Sensor fusion technology is becoming
nominal trajectory without taking into account vehicle
increasingly popular for agricultural navigation[12,20,25].
dynamics[10,11,13,21]. Some of the research showed very
Zhang et al.[9] developed an on-field navigation
good accuracy of less than 5 cm not only on a straight
system using a vision sensor, FOG and RTK GPS.
line but also on a curved path as soon as the vehicle
Figure 8 shows the comparison results.
satisfied the pure rolling constrains.
Unfortunately, pure
[73]
al.
Garaca-Prez et
developed a hybrid agent for behavior architecture
rolling constraints are almost impossible to satisfy during
adapted to agricultural navigation.
agricultural tasks due to sliding, deformed tires or change
was equipped with several positioning sensors (DGPS,
in wheel-ground contact conditions, which degrade the
digital compass and dead-reckoning system) and safety
performance and stability of automatic guidance. Some
sensors
literature is related to improved kinematic models that
emergency stops) as well as an on-board processor,
can adapt to the sliding influence and promise guidance
wireless
accuracy (e.g., Lenain et al.[74,75]; Fang et al.[76]). The
electrohydraulic actuators.
sliding effects have been taken into account for trajectory
were proposed to overcome the absence of GPS signals
tracking control of agricultural vehicles and three
so as to obtain continuous and precise positioning.
variables
characterizing
the
sliding
effects
(laser
rangefinder,
communication
The farming vehicle
bumper,
system
inclinometers,
(WLAN)
and
Sensor-fusion algorithms
were
introduced into the kinematic model based on geometric

and velocity constraints. An ideal refined kinematic
model was obtained in which sliding effects appeared as
additive
unknown
parameters
using
linearized
approximation.
3) Dynamic model
Dynamic models are fairly complex for agricultural
vehicle navigation, since describing all vehicle features
(e.g., inertia, sliding, springing) leads to very large,
Figure 8 Comparison of navigation accuracy in vehicle offset

from the desired path using sensor fusion, vision only, and
GPS-FOG only based navigation controls [9]
intricate models. In particular, most of the parameter

values (mass, wheel-ground contact conditions, tire and
5) Neural steer model
wheel deformation) are difficult to obtain even based on
Noguchi and Terao [20] designed a neural network (NN)
experimental identification. However, agricultural vehicle
vehicle controller in which the motion of the mobile
tasks involve mostly dynamic processing and researchers
agricultural robot was specified as a nonlinear system
are interested in investigating this
[77-79]
with high learning ability.
This NN model was applied
4) Sensor fusion
to navigation on an asphalt surface, with an accuracy of
The principle of sensor fusion is to combine
0.08 m in the offset.
information from various sensing sources (e.g., GPS and
Noguchi et al.
[58]
used an NN
model to correct the geomagnetic direction sensor for the
September, 2009
Vol. 2 No.3
11
A field test was conducted on
developed and implemented in guidance systems [9, 82,83].
a square path (40-m sides) in a meadow. The maximum
OConnor et al.[13] used a steering controller based on a
directional angle error was 14 using the conventional
set of linear motion equations.
inclination of the vehicle.
method, but only 1 using the NN.
Zhu et al.
[80]
Inoue et al. [84] developed
an adaptive steering controller that corrects the steering

Cho and Lee[85] used a fuzzy logic
designed an NN vehicle model for estimating vehicle
system delay.
behavior on sloping terrain. Bernoullis lemniscate was
controller for the autonomous operation of an orchard
employed to acquire training pairs. Genetic algorithms
speed sprayer.
and back propagation algorithms were used to train the
steering controller and obtained good curved-path
NN vehicle model.
guidance results.
The tractor was successfully guided
Kise et al.[43] developed an optimal

Zhang et al.[10] put forward a
along a predetermined path with mean and standard
kinematic model in which the steering linkage geometry
lateral deviation of 5 and 6.7 mm, respectively.
provided the gain between the hydraulic actuator and the
Ryerson
and Zhang[81] chose genetic algorithms to plan the
front wheels.
optimal path for a guided vehicle to avoid known
steering control loop based on the feedback signal from
obstacles.
the hydraulic steering actuator rather than from the front
wheels.
Steering controller
The system model was used to close the
Lenain et al.[75] considered agricultural vehicle
sliding and pseudo-sliding on slippery ground and used

A good control system is necessary irrespective of the
guidance sensor.
predictive model control to preserve accuracy.
The controller translates sensor
position deviation signals into a voltage signal that is

used to open a valve forcing the hydraulic ram in the
steering circuit to change the steering angle of the front or
An actuator, combined with the vehicle status, was

used to convert the control signal from a feedback
controller to the appropriate mechanical adjustment in
steering angle to provide the position of the vehicle.
rear axle, or, in the case of side shifting the equipment, an

additional ram to adjust the position of the equipment
Discussion
Since the time when the first driverless tractor
relative to the tractor or the row of plants.

Agricultural vehicles often work on different types of
prototype was created 50 years ago (Morgan 1958),
terrain, even and uneven, or changing and unpredictable
research into automatic guidance has steadily progressed,
terrain ranging from asphalt to spongy topsoil in the field.
particularly in the case of guidance system technologies,
In the case of automatic or autonomous navigation,
which have improved remarkably in the last two decades.
steering controllers should be able to provide appropriate
However, with the exception of GPS receivers, vision
steering action in response to the variation in equipment
sensors, laser rangers, gyroscopes and GDS, the
operation state, traveling speed, tire cornering stiffness,
commercialization of prototype agricultural guidance
ground
parameters
systems is very low. Various reasons are behind the
Consequently, steering
absence of funds for developing these prototypes into
conditions,
and
many
influencing steering dynamics.
other
controller design for agricultural vehicles is a difficult
commercial products.
challenge.
disuse as society has developed.
Some cases have fallen into

For example, new
Most modern agricultural vehicles employ some form
technology or production causes the prototype market to
of hydraulic steering system, and recent developments in
devalue, and performance standards for environmental
automatic
advanced
protection and implementing tractors are changing.
system
Nevertheless, some general conclusions can be drawn
considering vehicle dynamics, such as terrain conditions
regarding the failure of many prototype service robots
and vehicle status (speed and/or acceleration).
to reach commercial viability.
steering
modifications
to
controllers
the
existing
include
hydraulic
Various
steering controllers, including PID, feed-forward PID
The environmental and performance requirements for
(FPID), and fuzzy logic (FL) controllers, have been
agricultural vehicle guidance operation are extremely
12
September, 2009
strict (see Section 2).
Vol. 2 No.3
In addition to this barrier (a more
declining prices, it seems inevitable that these two
difficult consideration than other guidance applications),
technologies will be fused together or one of them will
there are others barriers that have not yet been resolved
be
from many years ago.
[24]
Hague et al.
concluded that
fused
[19,61]
gyroscopy
with
another
[11]
, GDC
technology,
or laser radar
[14]
such
as
, to realize
dead-reckoning sensors lead to the accumulation of errors
autonomous vehicles in agriculture, allowing real-time
resulting in positional drift; laser or radar and image
image processing with a digital controller on a simple PC,
based artificial landmark positioning systems are a direct
precision positioning with an RTK DGPS system or
function of positioning, and not prone to accumulating
heading computation with a traditional gyroscope.
drift errors, but the beacons take time to set up and may
However, if the guidance system for agriculture is
result in ambiguous and unreliable results due to false
commercialized, the following product research will be
detection and failure to detect obscured beacons.
needed as single technologies mature.
The
popular machine vision and GPS also have their

respective advantages and disadvantages.
Machine
An integrated
consideration may be better.

1) Evaluation of economic feasibility.
Electronics,
vision is an inexpensive and passive sensor, which has
computers, sensors and attachments are declining in price,
some excellent computer algorithms to support and
mostly because high commercial demand enabled their
[14,25]
. However, it also has
manufacture at great economies of scale; however, the
difficulty dealing with changing light conditions,
cost of designing and producing the special-purpose parts
shadows, direct sunlight and other difficulties with
for agricultural guidance systems will increase markedly.
extracting guidelines from the images captured in the
An evaluation of economic feasibility is necessary to
working environment.
determine the market value and understand the difference
advance successful research
Although most problems can be
solved with electronic shutters, automatic diaphragms,

color differences and the right position and adjustment of
compared with old conventional systems.

2) Improvement of robustness in versatility and
the camera, a row of plants or a furrow is needed to guide
dependability
the vehicle using image processing, and tasks such as
agricultural machine operates in a harsher environment
spraying or fertilizing uncultivated fields need another
(often in paddy fields), but its operating time adaptability
strategy. GPS is different than machine vision, as it is not
is stronger than other machines for harvesting, sowing
affected by environmental variations, and real-time
and spraying and the operating times are usually pivotally
kinematic (RTK) GPS can provide better accuracy.
related to benefit the farmers.
Nevertheless, GPS sensor accuracy depends on the
should be sufficiently robust to work effectively under
position of the satellites.
varying conditions.
In urban environments,
especially in narrow streets (urban canyons), buildings

can occult the microwaves from satellites. Moreover, a
of
mechanical
technology.
The
Hence, such machines
Todays technologies have not
always proven capable of delivering this performance.

3) Addition of safety to avoid the risk of damage
system using GPS for guidance requires that a
resulting from the use of such machines.
predetermined path be given for the vehicle to follow.
safety standard is different from countries, the use of
Secondly, a kinematic GPS is very expensive. GPS
automated machine will substantially increase a users
guidance systems pose a problem in terms of positioning
legal liability.
the antenna on the roof of the agricultural vehicle with the
all damages resulting from the uses of such machines, the
equipment working at the ground level.
cost may be very expensive.
This means that
Although the
If the users utilize the insurance against

Thereby, it should be
on sloping ground and with changing soil conditions,
necessary to consider the safety of such machines based
deviations can occur between the virtual guideline and the
on different safety standard.
path described by the equipment. Solving this problem

requires attitude measurement.
With the advent of computer vision and GPS and their
4) Provision of service system.
A sophisticated
machine that is broken cannot do a better job than a

primitive
machine.
As
machines
become
more
September, 2009
complicated, the skills required for their operation and

maintenance increase in proportion.
Vol. 2 No.3
13
in the future.
For agricultural
application, the service system will be more important
[References]
than in a factory environment, because there are very few,

if any, technicians who will have the expertise or
[1]
2006; 50(4): 8394. (in Japanese).
equipment for repairs and maintenance.

With the emergence of new technologies in the
industry, research into their application to agricultural
[2]
Willrodt F L. Steering attachment for tractors. 1924; U.S.
[3]
Sissons R. Plowing in circles saves time. The Prairie Farmer,
[4]
Morgan K E. A step towards an automatic tractor. Farm
Patent No. 1506706.
vehicle guidance systems will contribute to the realization

of autonomous agricultural vehicles or robots in the
future.
1939; 111(20): 7.
For example, omnidirectional vision sensors[86]
have become increasingly attractive for autonomous

navigation systems.
Mech, 1958; 10(13): 440441.

[5]
captured by the sensor are an orthographic projection of
399.
[6]
Reid J F, Zhang Q, Noguchi N, Dickson M. Agricultural

automatic guidance research in North America. Computers
The images (obtained without
rotating the robot) are a 360 view of the environment
Harrell R C, Adsit P D, Pool T A, Hoffman R. The Florida

Robotic Grove Lab. ASAE Transactions, 1990; 33: 391
The camera and mirror are
mounted at the top of the mobile robots platform. Images

the ground plane.
Tamaki K. Agriculture and Robots. Tokyo Univ of Agric,
and Electronics in Agriculture, 2000; 25: 155167.

[7]
Tillett N D. Automatic guidance sensors for agricultural field
and therefore are not sensitive to wheel slippage or small
machines: a review. Journal of Agricultural Engineering
vibrations.
Research, 1991; 50(33): 167187.
enough
This low-cost sensing device provides
information
for
our
navigation
system.
[8]
Murakami N, Ito A, Will J D, Inoue K, Kita K, Miyarua S.
Although it is not easy to obtain distance estimations
Environment
from an omnidirectional image due to the shape of the
Omni-vision and image map. In: Proceedings of the 3rd
We proposed this system as a potential substitute for the
technique
using
hyper
IFAC Intl. Workshop Bio-Robotics, Sapporo, Japan, 2006; p.
mirror, the apparent angles of objects around the robot are

relatively accurate and easy to derive from the image[87,88].
identification
317320.
[9]
Zhang Q, Reid J F, Noguchi N. Agricultural vehicle

navigation using multiple guidance sensors. In: Proceedings
GPS function for localization using landmarks in the
of International Conference on Field and Service Robotics.
working environment.
Pittsburgh, PA, USA, 1999; p. 293298.
[10] Zhang Q, Wu D, Reid J F, Benson E R. Model recognition
Conclusions
and validation for an off-road vehicle electrohydraulic
This paper provides a brief review of the research on

guidance system technologies in agricultural vehicles
over the past 20 years.
Although the research
developments are abundant, there are some shortcomings

(e.g., low robustness of versatility and dependability of
steering controller. Mechatronics, 2002; 12(6): 845858.

[11] Benson E, Stombaugh T, Noguchi N, Will J, Reid J F. An
evaluation of a geomagnetic direction sensor for vehicle
guidance in precision agriculture applications. ASAE Paper
983203. ASAE. St. Joseph, MI. 1998.
[12] Benson E, Reid J F, Zhang Q. Machine vision-based
technologies) that are delaying the improvements
guidance system for agricultural grain harvesters using
required for commercialization of the guidance systems.
cut-edge detection. Biosystems Engineering, 2003; 86(4):389
It can be concluded that either GPS and machine vision
398.
technologies will be fused together or one of them will
[13] OConnor M, Bell T, Elkaim G, Parkinson B. Automatic
be fused with another technology (e.g., laser radar) as
Steering of Farm Vehicles Using GPS. Paper presented at 3rd
the trend development for agricultural vehicle guidance
International Conference on Precision Agriculture, ASA,
systems.
The application of new popular robotic
technologies for agricultural guidance systems will

augment the realization of agricultural vehicle automation
CSSA, and SSSA, Madison, WI. 1996.

[14] Subramanian V, Burks T F, Arroyo A A. Development of
machine vision and laser radar based autonomous vehicle
guidance systems for citrus grove navigation. Comput
14
September, 2009
Electron Agric, 2006; 53(2): 130143.
Vol. 2 No.3
12.
[15] strand B. Vision based perception or mechatronic weed
[29] Pinto F A C, Reid J F, Zhang Q, Noguchi N. Vehicle
control. Doctor of Philosophy Thesis, Chalmers and
guidance parameter determination from crop row images
Halmstad Universities, Sweden, 2005.
using principal component analysis. Journal of Agricultural
[16] strand B, Baerveldt A J. A vision based row-following

system for agricultural machinery. Mechatronics, 2005;
15(2): 251269.
Engineering Research, 2000; 75(3): 257264.

[30] Okamoto H, Hamada K, Kataoka T, Terawaki M, Hata S.
Automatic guidance system with crop row sensor. In:
[17] Sgaard H T, Olsen H J. Determination of crop rows by
Proceedings of Conference on Automation Technology for
image analysis without segmentation. Comput Electron Agric,
Off-Road Equipment (Chicago, Illinois, USA) 701P0502, p.
2003; 2(38): 141158.
307316. 2002.
[18] Imou K, Okamoto T, Torii T, Ishida M, Takanaga H, Okado
[31] Tillett N D, Hugue T. Computer-vision-based hoe guidance
A, Itokawa N. Autonomous Tractor for Forage Production.
for cereals an initial trial. Journal of Agricultural
In: Proc. 13th International Congress on Agricultural

Engineering, 1998; 3: 361368.
[19] Nagasaka Y, Umeda N, Kanetai Y, Taniwaki K, Sasaki Y.
Automated rice transplanter using global positioning and
Engineering Research, 1999; 74(3): 225236.

[32] Kise M, Zhang Q, Rovira M F. A stereovision-based crop
row detection method for tractor-automated guidance.
Biosystems Engineering, 2005; 90(4): 357367.
gyroscopes. Comput Electron Agric, 2004; 43 (3): 223234.
[33] Kaizu Y, Imou K. A dual-spectral camera system for paddy
[20] Yukumoto O, Matsuo Y, Noguchi N. Navigation technology
rice seedling row detection. Comput. Electron Agric, 2008;
for tilling robots. Robotisation of agricultural vehicles and

the outline of navigation system for tilling robot. Proc.
International Symposium on Mobile Agricultural Bus-System
Lab and PA for the Large Scale Farm Mechanization,
Sapporo, Japanese Society of Agricultural Machinery, 1997;
p.5978.
[21] Noguchi N, Terao H. Path planning of an agricultural mobile
robot by neural network and genetic algorithm. Comput
Electron Agric, 1997; 18(2-3): 187204.
[22] Noguchi N, Kise M, Ishii K, Terao H. Field automation using
robot tractor. In: Proceedings of Conference on Automation
Technology for Off-Road Equipment, ASAE, Chicago, IL.
p. 239245. 2002.
[23] Billingsley J, Schoenfisch M. The successful development of
a vision guidance system for agriculture. Computers and
Electronics in Agriculture, 1997; 16(2): 147163.
63(1): 4956.
[34] Larsen W E, Nielsen G A, Tyler D A. Precision navigation
with GPS. Comput Electron Agric, 1994; 11(1): 8595.
[35] Gan-Mor S, Ronen B, Josef S, Bilanki Y. Guidance of
autonomous vehicle for greenhouse transportation, Acta
Hortic, 1997; 443: 99104.
[36] Bell T. Automatic tractor guidance using carrier-phase
differential GPS. Comput. Electron Agric, 2000; 25(1-2): 53
66.
[37] Yukumoto O, Matsuo Y, Noguchi N. Robotization of
agricultural vehicles description of the tilling robot. Journal
of Agricultural Engineering Research, 2000; 34: 107114.
[38] Abidine A Z, Heidman B C, Upadhyaya S K Hills D J.
Application of RTK GPS based auto-guidance system in
agricultural production. ASAE Paper No. O21152. ASAE,
St. Joseph, MI. 2002.
[24] Hague T, Marchant J A, Tillet N D. Ground based sensing
[39] Gan-Mor S, Upchurch B L, Clark R L, Hardage D.
systems for autonomous agricultural vehicles, Computers and
Implement guidance error as affected by field conditions
Electronics in Agriculture, 2000; 25(1-2): 1128.
using automatic DGPS tractor guidance. ASAE Paper No.
[25] Han S, Zhang Q, Ni B, Reid J F. A guidance directrix

approach to vision-based vehicle guidance systems. Comput
Electron Agric, 2004; 3(43): 179197.
[26] Marchant J A, Brivot R. Real-time tracking of plant rows
using a Hough transform. Real-Time Imaging, 1995; 1(5):
36371.
021153. ASAE, St. Joseph, MI. 2002.

[40] Slaughter D C, Giles D K, Downery D. Autonomous robotic
weed control systems: A review. Comput. Electron Agric,
2008; 61(1): 6378.
[41] Stoll A, Kutzbach H D. Guidance of a Forage Harvester with
GPS. Precision Agriculture, 2000; 2: 281291.
[27] Marchant J A. Tracking of row structure in three crops using
[42] Kise M, Noguchi N, Ishii K, Terao H. Development of
image analysis. Comput Electron Agric, 1997; 15(2): 161
agricultural autonomous tractor with an RTK-GPS and a
79.
FOG. Proceedings of the Fourth IFAC Symposium on
[28] Hague T, Tillett N D. A bandpass filter-based approach to

crop row location and tracking. Mechatronics, 2001; 11(1): 1
Intelligent Autonomous Vehicles, Sapporo, Japan, p. 99

104. 2001.
September, 2009
[43] Kise M, Noguchi N, Ishii K, Terao H. The development of

the autonomous tractor with steering controller applied by
Vol. 2 No.3
15
for Mobile Robots IEEE Transactions on Robotics and

Automation, 1995; 1(3): 328342.
optimal control. Proceedings of Conference on Automation
[55] Schnberg T, Ojala M, Suomela J, Torpo A, Halme A.
Technology for Off-Road Equipment, Chicago, USA, p. 367
Positioning an autonomous off-road vehicle by using fused
373. 2002.
DGPS and inertial navigation. International Journal of
[44] Ehsani M R, Sullivan M D, Zimmerman T L, Stombaugh T.

Evaluating the dynamic accuracy of low-cost GPS receivers.
Systems Science, 1996; 27(8): 745752.

[56] Yu B, Jain A K. Lane boundary detection using a
In: Proceedings of the ASAE Annual International Meeting,
multiresolution
Nevada, USA. 2003.
International Conference on Image Processing, 1997; 2: 748
[45] Borenstein J. Experimental Results from Internal Odometry

Error Correction with the OmniMate Mobile Platform. IEEE
Transactions on Robotics and Automation, 1996; 14(6): 963
969.
[46] Chenavier F, Crowley J. Position Estimation for a Mobile
Robot Using Vision and Odometry. Proceedings of IEEE
International Conference on Robotics and Automation, Nice,
France, May 1214, pp. 25882593. 1992.
[47] Imou K, Okamoto T, Kaizu Y, Yoshii H. Ultrasonic Doppler
Speed Sensor for Autonomous Vehicles. Journal of the
Japanese Society of Agricultural Machinery, 2001; 63(2): 39
46.
[48] Imou K, Okamoto T, Kaizu Y, Ishida M, Sawamura A,
Sumida Y. Ultrasonic Doppler sensor for measuring vehicle
speed in forward and reverse motions including low speed
motions. Agricultural Engineering International: CIGR
Journal of Scientific Research and Development, Vol. III.
2001.
[49] Holmqvist R. A laser-optic navigation system for driverless
vehicle, Arnex Navigation AB, R. Holmqvist, ARNBL
344:15. 1993.
[50] Ahamed T, Takigawa T, Koike M, Honma T, Yoda A,
Hough
transform.
Proceedings
of
51.
[57] Ishida M, Imou K, Okado A, Takenaga H, Honda Y, Itokawa
N, Shibuya Y. Autonomous tractor for forage production.
Journal of the Japanese Society of Agricultural Machinery,
1998; 60(2): 5966. (in Japanese).
[58] Noguchil N, Ishii K, Terao H. Development of an agricultural
mobile robot using a geomagnetic direction sensor and image
sensors. Journal of Agricultural Engineering Research, 1997;
67: 115.
[59] Harper N, McKerrow P. Recognising plants with ultrasonic
sensing
for
mobile
robot
navigation.
Robotics and
Autonomous Systems, 2001; 34: 7182.

[60] Yekutieli O, Pegna F G. Automatic guidance of a tractor in a
vineyard. In: Proceedings of Conference on Automation
Technology for Off-Road Equipment, Illinois, USA. 2002.
[61] Morimoto E, Suguri M, Umeda M. Vision-based navigation
system for autonomous transportation vehicle. Precision
Agriculture, 2005; 6: 239254.
[62] Lee J W, Choi S U, Lee Y J, Lee K S. A study on recognition
of lane and movement of vehicles for port AGV vision
system. In: Proceedings of the 40th SICE Annual Conference.
Hasegawa H, Junyusen P, Zhang Q. Characterization of laser
International Session Papers, 2527 July, pp. 3841. 2001.
range finder for in-field navigation of autonomous tractor. In:
[63] Leemans V, Destain M F. Line cluster detection using a
Proceedings of Conference on Automation Technology for
variant of the Hough Transform for culture row localization.
Off-Road Equipment, Kyoto, Japan. 2004.
Image and Vision Computing, 2006; 24(5): 541550.
[51] Junya T, Kiyoshi T, Yutaka T, Itashi I, Masayoshi K, Koji T.
[64] Kalman R E. A new approach to linear filtering and
2-D localization of autonomous farm vehicle using laser
prediction problem. Trans. ASME: J. Basic Eng. 1960; 82
distance sensor. Journal of the Japanese Society of
(March): 3540.
Agricultural Machinery, 2003; 67(5): 4553.

[52] Barawid J O C, Mizushima A, Ishii K, Noguchi N.
Development of an autonomous navigation system using a
two-dimensional laser scanner in an orchard application.
[65] Bergeijk J V, Goense D, Keesman K J, Speelman L. Digital

filters to integrate global positioning system and dead
reckoning. J Agric Eng Res, 1998; 70(2): 135143.
[66] Kubo Y, Kindo T, Ito A, Sugimoto S. DGPS/INS/wheel
sensor integration for high-accuracy land-vehicle positioning.
[53] Tofael A, Tomohiro T, Masayuki K, Tsuyoshi H, Hideo H,
In Proc. of 12th International Technical Meeting of the
Zhang Q. Navigation Using a Laser Range Finder for
Satellite Division of the Institute of Navigation, p. 555564.
Autonomous Tractor (Part 1) Positioning of Implement.
ION GPS99. Fairfax, Va.: The Institute of Navigation.
Journal of the Japanese Society of Agricultural Machinery,

2006; 68(1): 6877.
[54] Barshan B., Durrant-Whyte H F. Inertial Navigation Systems
1999.
[67] Madhukar B R, Nayak R A, Ray J K, Shenoy M R. GPS/DR
integration using low-cost sensors. In Proc. of 12th
16
September, 2009
Vol. 2 No.3
International Technical Meeting of the Satellite Division of
navigation applications. IMTC 2005 Instrumentation and
the Institute of Navigation, p. 537544. ION GPS99.
Measurement Technology Conference, Ottawa, Canada, 17
Fairfax, Va.: The Institute of Navigation. 1999.
19 May, p. 13301334. 2005.
[68] Nrremark M, Griepentrog H W, Nielsen J, Sgaard H T.
[79] Bouton N, Lenain R, Thuilot B, Martinet P. Backstepping
The development and assessment of the accuracy of an
observer dedicated to tire cornering stiffness estimation
autonomous GPS-based system for intra-row mechanical
application to an all terrain vehicle and a farm tractor.
weed control in row crops. Biosystems Engineering, 2008;
Proceedings of the 2007 IEEE/RSJ International Conference
101(4): 396410.
on Intelligent Robots and Systems, San Diego, CA, USA, Oct
[69] Rudolph V D M, Eric A W. Sigma-Point Kalman Filters for
29 Nov 2, p. 17631768. 2007.
Integrated Navigation. In: Proceedings of the 60th Annual
[80] Zhu Z X, Torisu R, Takeda J I, Mao E R, Zhang Q. Neural
Meeting of the Institute of Navigation (ION), Dayton, OH,
network for estimating vehicle behavior on sloping terrain.
June 2004.
[70] Zhang Y, Gao L, Tian L. INS/GPS integrated navigation for

wheeled agricultural robot based on sigma-point Kalman
Filter. Asia Simulation Conference 7th Intl. Conf. on Sys.
Simulation and Scientific Computing. p. 14251431. 2008.
[71] Freeland S R, Wilkerson B J, Hart E W. Instrumentation for
measurement of tractor yaw. ASAE Paper No. 92-3552.
ASAE, St. Joseph, MI. 1992.
field coverage using genetic algorithms. Agricultural

Engineering International: the CIGR Ejournal. Manuscript
ATOE 07014. Vol. IX (July). 2007.
[82] Qiu H,
Zhang
Q, Reid
F,
Wu
D. Nonlinear
feed-forward-plus-PID control for electrohydraulic steering

systems. In Fluid Power Systems and Technology 6: 8994.
[72] Kodagoda K R S, Wijesoma W S, Teoh E K. Fuzzy speed

and steering control of an AGV. IEEE Trans. Control Syst.
Technol, 2002; 10 (1): 112120.
New York, N.Y.: ASME, 1999.

[83] Wu D, Zhang Q, Reid J F. Adaptive steering controller using
a Kalman estimator for wheel-type agricultural tractor.
[73] Garaca-Prez L, Garaca-Alegre M C, Ribeiro A, Guinea D.

An agent of behavior architecture for unmanned control of a
farming vehicle. Comput. Electron Agric, 2008; 60(1): 39
48.
Robotica, 2001; 19: 527533.

[84] Inoue K, Otuka K, Sugimoto M, Murakami N. Estimation of
place of tractor and adaptive control method of autonomous
tractor using INS and GPS, Preprints of the International
[74] Lenain R, Thuilot B, Cariou C, Martinet P. Model predictive

control for vehicle guidance in the presence of sliding:
application to farm vehicles path tracking. International
Conference on Robotics and Automation, Barcelona, Spain, p.
897902. 2005.
Workshop on Robotics and Automated Machinery for

Bio-Productions, EurAgEng, Gandia, p. 27736. 1997.
[85] Cho S I, Lee J H. Autonomous speed sprayer using
differential global positioning system, genetic algorithm and
fuzzy control. Journal of Agriculture Engineering and
[75] Lenain R, Thuilot B, Cariou C, Martinet P. High accuracy

path tracking for vehicles in the presence of sliding:
application to farm vehicle automatic guidance
[81] Ryerson A E F, Zhang Q. Vehicle path planning for complete
for
agricultural task. Auton Robot, 2006; 21: 7997.

[76] Fang H, Ruixia F, Thuilot B, Martine P. Trajectory tracking
control of farm vehicles in the presence of sliding. Robotics
and Autonomous Systems. 2006; 54(10): 828839.
[77] Bevly D M, Gerdes J C, Parkinson B W. A new yaw dynamic
model for improved high speed control of a farm tractor.
Journal of Dynamic System, Measurement, and Control,
2002; 124(4): 659667.
[78] Feng L, He Y, Bao Y, Fang H. Development of trajectory
model for a tractor-implement system for automated
Research, 2000; 76: 111119.

[86] Ayako T, Yoshinobu A, Makoto M. Autonomous agricultural
robot with seeding function self localization using
Omni-directional vision. Nippon Kikai Gakkai Robotikusu,
Mekatoronikusu Koenkai Koen Ronbunshu (CD-ROM),
2A1-B13. 2006 (in Japanese).
[87] Imou K, Tani S, Yokoyama S. Position detection of vehicle
using an omni-directional camera. In Proc. ISMAB, Taiwan.
2008.
[88] Li M, Imou K, Tani S, and Yokoyama S. Localization of
agricultural
vehicle
using
landmarks
based
on
omni-directional vision. In: Proc. ICCEA, Manila. p.140

144. 2009.
September, 2009
Vol. 2 No.3
17
Developing a modeling tool for flow profiling in irrigation

distribution networks
Yanbo Huang1, Guy Fipps2
(1. Application and Production Technology Research Unit, US Department of Agriculture,
Agricultural Research Service (USDA-ARS), Stoneville, MS 38776, USA;
2. Department of Biological and Agricultural Engineering, Texas A&M University, College Station, TX 77843, USA)
Abstract: Efforts are underway to rehabilitate the irrigation districts, such as in the Rio Grande Basin in Texas.
Water
distribution network models are needed to help prioritize and analyze various rehabilitation options, as well as to
scientifically quantify irrigation water demands, usages, and losses, and to help manage gate automation. However,
commercially available software packages were limited in applications due to their high cost and operational difficulty.
This study aims to develop a modeling tool for modeling the water flow profile in irrigation distribution networks. The
goal of developing the modeling tool was to make the modeling process simple, fast, reliable and accurate. On the basis
of methodological study, the modeling tool has been developed for branching canal networks with the assumption of
steady gradually varied flow. The flow profile calculation of the tool was verified from a single channel with 1% root
mean squared error compared to the benchmark calculation and a branching network with 5% to 12% relative errors
compared to check point measurement along the network. The developed modeling tool will be able to play an
important role in water quantification for planning, analysis and development for modernization of irrigation systems.
Keywords: irrigation distribution network, modeling tool, flow profiling
DOI: 10.3965/j.issn.1934-6344.2009.03.017-026
Citation: Yanbo Huang, Guy Fipps. Developing a modeling tool for flow profiling in irrigation distribution networks.
Int J Agric & Biol Eng, 2009; 2(3): 1726.
and losses, and to help manage gate automation.
Introduction
There
has been much research in developing computer models
Irrigation distribution networks are used extensively
and software packages for water resources planning and
Irrigation districts deliver
management through the past three decades[1]. Models
water to farms through the channels and pipelines. Efforts
and software packages are commercially or research
are underway to rehabilitate the irrigation districts.
available for flow modeling and gate automation of
Quantitative evaluation tools are needed to help prioritize
irrigation channels.
and analyze various rehabilitation options, as well as to
Hydraulics, Delft, Netherlands), an integrated 1D/2D
scientifically quantify irrigation water demands, usages,
modeling program for water management, design,
for agricultural water supply.
Received date: 2009-02-19
Accepted date: 2009-09-10
Biographies: Yanbo Huang, Ph.D., Agricultural Engineer,

USDA-ARS- APTRU, 141 Experiment Station Rd., Stoneville, MS
38776, Tel: (662)820-7715. Email: Yanbo.Huang@ars.usda.gov;
Guy Fipps, Ph. D, P.E. Professor and Extension Agricultural
Engineer, Department of Biological and Agricultural Engineering,
Texas A&M University, College Station, TX 77843, Email:
g-fipps@tamu.edu
Examples are: SOBEK (Delft
planning and policy making in river, rural and urban

systems
CanalCAD
Grenoble,
(http://www.sobek.nl/prod/index.html);
(Laboratoire
France;
dHydraulique
Parrish Engineering,
de
France,
Beaverton,
Oregon, USA), a hydrodynamic simulator of both steady

and unsteady flow in canal systems with manual or
automatic
gates
(http://www.iihr.uiowa.edu/projects/
18
September, 2009
Vol. 2 No.3
Hydraulic
discretization of the Saint-Venant equations and solves
Institute, Hrsholm, Denmark), a versatile and modular
the resulting equations using the sparse matrix solution
engineering software tool for modeling conditions in
technique.
rivers, lakes/reservoirs, irrigation canals and other inland
in a series of linearly connected reaches, and branched as
water
well as looped canal networks.
canalcad/index.html);
systems
Mike
11
(Danish
(http://www.dhisoftware.com/mike11);
The model is applicable for simulating flow

In general, unsteady
SIC (Cemagref, Antony Cedex, France), a simulation
gradually varied flow (USGVF) can be described by the
model
Saint-Venant
for
canal
automation
design (http://canari.
montpellier.cemagref.fr/papers/sic30.pdf);
HEC-RAS
equations[6].
These
equations
are
simultaneous partial differential equations with a number
(IWR, US Army Corps of Engineers, Davis, California,
of boundary conditions.
USA), a software package that allows one-dimensional
unsteady canal model requires serious investments of
steady and unsteady flow calculations in natural channels
time and personnel[7].
However, in practice use of a

As a special case of USGFV,
and
steady gradually varied flow (SGVF) can be described by
CanalMan (Utah State University, Logan, Utah, USA) a
a single ordinary differential equation[6], which is much
model that performs hydraulic simulations of unsteady
more
flow
equations.
(http://www.hec.usace.army.mil/software/hec-ras);
in
branching
canal
networks
(http://www.
engineering.usu.edu/bie/software/canalman.php).
easily
implemented
than
the
Saint-Venant
In many cases the description of SGVF is
These
very useful and effective and the USGVF could be
models or software packages are for general use and
simplified to cascaded SGVFs in solving problems in
either expensive, such as SOBEK and CanalCAD or are
flow computation and analysis.
difficult to be customized for applications under specific
The objective of the study was to develop a modeling
conditions even free downloadable, such as HEC-RAS
tool based on the description of SGVF for modeling the
and CanalMan.
water flow profile in irrigation distribution networks in
Models have been evaluated for irrigation systems.
the Rio Grande Basin in Texas and other similar areas.
Wallender[2] has done model simulation for both a single
The developed modeling tool will make the modeling
furrow as well as on a field-wide basis.
process simple, fast, reliable and accurate.
Model
simulations were evaluated to determine the importance

to irrigation performance of each spatially-varying model
Study area
input. Esfandiari and Maheshwari[3] studied four furrow
Irrigated lands in different areas have different
irrigation models, referred to as the Ross, Walker,
characteristics. This study will focus on the irrigated
Strelkoff and Elliott models for their prediction of
areas with the following characteristics:
advance and recession times and runoff, and for their
The waterways are shallow and have small hydraulic
computational time per simulation run and volume
gradients.
balance error under three field conditions in south-east
is small and the water flows mildly from upstream to
Australia.
[4]
Hidalgo et al.
developed a procedure for
calibrating on-demand irrigation network models.

more commonly used objective functions.
downstream
This
procedure compared a new objective function with two
with gravity and sufficient head
pressure;
This
procedure was applied to an on-demand irrigation
In other words, the channel bottom slope
The distribution networks are dendritical, i.e. the

routes of the networks are branched but not looped.
The networks are open channels
network located in Tarazona de La Mancha (Albacete,
In the Lower Rio Grande Valley in Texas (Figure 1),
Spain) where flow and pressure at hydrant level was
the project area, the elevations range from sea level in the
measured.
[5]
Islam et al.
presented a hydraulic
east to about 200 m in the northwest, but are mainly less
simulation model developed for steady and unsteady flow
than 100 m.
simulation in irrigation canal network.
ways are shallow and have low gradients. The canals
the
implicit
four-point
Preissmann
The model uses

scheme
for
Much of the area is nearly level. Drainage
and pipelines in the distribution networks have small
September, 2009
Developing a modeling tool for flow profiling in irrigation distribution networks
hydraulic gradients with few relief pumps. The objects
Vol. 2 No.3
19
of this study will be irrigation canals.
Figure 1 Service areas of the irrigation districts in Lower Rio Grande Valley of Texas
layout of a distribution network under different initial and
Computing methods
boundary hydraulic conditions.

In the open-channels (canals) of irrigation networks,
water flows are typically categorized as:
Non-uniform flow is the prevailing flow conditions in

irrigation systems.
For the area the irrigation channels
1) Steady uniform flow (SUF); 2) Steady gradually
are shallow and have small hydraulic gradients such as
varied flow (SGVF); and 3) Unsteady gradually varied
the Rio Grande Basin in Texas the SGVF is the dominate
flow (USGVF).
flow type unless some transient processes typically
The SGVF can be computed and analyzed by
happened around gate structures would result in the
observing the conservation of mass and energy with an
USGVF flow condition.
ordinary differential equation[6].
the SGVF profiles in irrigation distribution networks is
Further, the USGVF
Therefore, the computation of
can be computed and analyzed using the Saint-Venant
the technique needed in developing the modeling tool.
equations observing the conservation of mass and
3.1
momentum[6].
It can be derived mathematically that the
SGVF is a special case of the USGVF.
SGVF flow profile computation

The computation of the SGVF profiles basically
The
solves the governing ordinary differential equation. The
Saint-Venant equations are partial differential, so the
main objective of the computation was to determine the
implementation of the computation is much more
shape of the flow profile. Broadly three methods of the
difficult. In practice the SGVF is very useful and
computation
effective in solving a lot of problems in flow computation
integration method, the direct-integration method, and the
and analysis.
step method.
With the fundamental equation the
solutions can be cascaded along a canal channel and the
were
classified
as[6]:
the
graphical-
The graphical-integration procedure is straightfor-
20
September, 2009
ward and easy to follow but it may become very laborious

when applied to actual problems.
Sf =
Because the
explicitly in terms of y for all types of channel cross

sections, a direct and exact integration of the equation is
developed either to solve the equation for a few special
cases or to introduce assumptions that make the equation
amenable to mathematical integration[6].
Basically a
respects, but no one has been found to be the best in all

applications.
This study gives a step method based on the need of
flow profile computation for irrigation channels.
This
method divides a channel to small reaches. The length

of the reaches cannot be too big because this may cause
the iterative procedure to fail, and cannot be too small
either because this should increase computational burden.
4/3
(3)
the upstream and downstream ends respectively; Ru and

Rd are the channel hydraulic radiuses of the upstream and
downstream ends respectively.
With the solutions as the initial conditions the
equation can be applied to the next reach and so on.
reach to the other. There are a great variety of step

methods. Some appear superior to others in certain
where Au and Ad are the channel cross section areas of
step method is to divide a channel into short reaches and

carry the computation step by step from one end of the
n2 Q2
A R
Au + Ad
A=
2
Ru + Rd
R=
2
differential equation of the SGVF cannot be expressed
practically impossible; hence, so far this method has been
Vol. 2 No.3
The computation at each reach is an iterative process.

Given Q, n, So, and channel cross section parameter such
as bottom width b and side slope s for a trapezoid cross
section, at the beginning the upstream end water depth yu
was set to be the downstream end water depth yd which
was from the solution of the previous reach or the initial
condition at the channel downstream end, i.e. yu=yd.
With the initial yu a new estimate of the unknown water
depth using equations (2) and (3) was calculated as yu .
With the divided reaches the computation starts from the
Then, the initial water depth was compared with the
downstream end of the channel for subcritical flow (from
estimated depth with y u yu < where was a pre-set
the upstream end for supercritical flow) by applying the
small number for stopping the iteration.
Bernoulli equation to the reach:
condition is met, the iteration will stop and y u is the
yu +
2
u
2
d
v
v
+ S o x = yd +
+ S f x
2g
2g
(1)
and downstream ends of the reach respectively; is the

velocity distribution coefficient which takes into account
that in channel cross-section the distribution of velocity is
not uniform;
solution; otherwise set yu = y u and continue the iteration.

3.2
where vu and vd are the flow velocities at the upstream
x is the length of the reach; So is the
channel bottom slope.
Branching network SGVF flow profiling

The algorithm above can be used to compute SGVF
flow profiling in a canal channel or a distribution network

by cascading the solutions step by step along the canal
channel and the layout of a distribution network under
different initial and boundary hydraulic conditions.
Branching
The solution of the equation for subcritical flow will
If the stopping
irrigation
distribution
dominated in the studied areas.
networks
are
This kind of networks
be water depth h and water level z=y+z at the upstream
typically consists of laterals, second-level laterals, and
end of the reach where z is the difference between the
even third-level laterals along a main canal. The flow
elevations at the upstream and downstream ends of the
profile computation over a branching network starts by
reach. Equation (1) can be reformed to solve the water
initializing discharge and water depth at the one end of
depth at the upstream end of the reach:
the main canal.
yu = yd + (
and
2
d
2
u
v
v
) S o x + S f x
2g 2g
(2)
Then when the computation proceeds to
a lateral, the computation needs to continue by initializing

discharge and water depth at the one end of the lateral.
When the computation proceeds to a second-level lateral,
September, 2009
the computation needs to continue by initializing

Keep on going like this until the
farm turnouts are reached and the computations needs to

recursively go back to the main canal.
The same
procedure follows when the second, third, laterals are

met.
The computation will stop when it proceeds to the
other end of the main canal.
Figure 2 shows the flow
Q = C fo LGo 2 g ( yu 0.5Go )
the water depth upstream of the gate structure; Cfo is the

discharge coefficient of the FO flow.
2) Submerged orifice (SO): it is submerged gated
flow
Q = Cso Lyd 2 g ( yu yd )
computation over a branching irrigation network. This
where
procedure can handle the branching irrigation networks in
structure;
arbitrary layouts as long as they only have the first-level
flow.
This procedure can be easily extended to the
(5)
yd is the water depth downstream of the gate

Cso is the discharge coefficient of the SO
3) Free non-orifice (FN): it is free weir flow

Q = Cnf L yu
cases of arbitrary branching networks with second-level,

third-level, and n-level laterals.
(4)
where L is the gate size; Go is the gate opening; yu is
chart of the procedure of subcritical SGVF profile
laterals.
21
1) Free orifice (FO): it is free gated flow
discharge and water depth at the one end of the

second-level lateral.
Vol. 2 No.3
(6)
where Cnf is the discharge coefficient of the FN flow.

4) Submerged non-orifice (SN): it is submerged weir
flow
Q = Csn Lyd 2 g ( yu yd )
(7)
where Csn is the discharge coefficient of the SN flow.

In practice, although water flow can transit from one
regime to the other, many canal gate structures and
channel constrictions such as flumes operate mostly
under a single flow regime.
To use any one of the equations (4), (5), (6), and (7)
to characterize the water flow though a specific gate
structure, the corresponding discharge coefficient, Cfo,
Cso, Cfn, or Csn, needs to be determined (calibrated).
The
[9]
calibration procedure is as follows :

1) Conduct field survey around the concerned gate
structure: gate dimensions and gate upstream and
downstream channel hydraulic characteristics.
2) Determine the flow regime by experience or by
some computation about water flow through the gate
Figure 2 Flow chart of computation of subcritical SGVF profile
over branch irrigation networks with first-level laterals
structure.
3) Find out the standard equation of a specific gate
structure for the determined flow regime: equation (4),
3.3
(5), (6), or (7).
Gate calibration
Gate is the most popular structure for controlling
water flow through irrigation channels.
In general, four
different flow regimes can occur at gate structures. Each

of the four regimes has a standard equation to
[9,10]
characterize the flow through the gate structure
4) Rearrange the equations (4), (6), and (7) in the

following general form:
Qp(yu, yd , Go , L) = Cq(yu, yd , Go)
where C is Cfo , Cfn, or Csn .
Equation (4) can be assumed in the form:
(8)
22
September, 2009
log10(Cso)=a+b*log10(yd/Go )
C++
(9)
programming
Vol. 2 No.3
language
was
chosen
for
where a and b are regression coefficients.
prototyping the modeling tool.
I. Based on the n sequential measurements of (Qi , yui,

ydi , Goi) (i=1,2,,n), calculate (qi (yui, ydi , Goi ), Qi pi (yui,
ydi , Goi , L)) for equation (8) or (log10(ydi /Goi ), log10(Csoi ))
designed and developed using the principles of OOP
for equation (9) (i=1,2,,n).

II. Based on the calculation, the regression equation is
(Object-Oriented Programming).
Model validation
The water flow profiles in a single irrigation canal
channel and a branching network irrigation scheme were
formulated as:
/Q
p = Cq
(10)
computed for model validation.
As the benchmark, the
data of the single irrigation canal channel were taken
for FO, FN, or SN flow where C is the estimated

value of C, or
from Chow (1959)[6].
The computing results were
compared with Chows computation to validate the

log10(Cso)= a + b *log10(yd/Go )
where
The programs were
(11)
a and b are the estimated values of a and b
computation of this study.

The irrigation scheme is a real-world irrigation
branching network (Figure 3), which spans about 1700 m
respectively.
III. The performance of the calibration can be
and is located at an irrigation district Jamaica around an
evaluated by calculating the standard deviations of
area that has similar characteristics as the Lower Rio
residuals.
Grande Valley of Texas.
The data were measured and
collected from the field survey and flow measurement.
Modeling tool prototyping
The computing results were verified with check point
Using the method above the modeling tool was
values along the irrigation system.
In the scheme, the
In order to make the
main canal goes through the points 1, 2, 3, 4, 5, 6, and 7.
modeling process simple, fast, and accurate, three
In the main canal at the upstream end is a sharp-crested
modules have been developed:
weir (HS1). A siphon wall is in the middle (HS2).
programmed and developed.
At
1) SGVF computation for a single canal channel
the downstream end is another sharp-crested weir (HS5).
2) SGVF computation for branching canal networks
Two laterals are from the main canal through two sluice
3) Flow computation through control sections
gates: HS3 and HS4 respectively.
The third module is for computing the flow through
down during field survey and measurement. HS4 was
gates, weirs, and flumes.
The discharge and depth
HS3 was fully shut
open to allow water flow to go through the points 5a, 5b,
relationships were calibrated and saved for model
5c, 5d, 5e, 5f, 5g and 5h.
implementation.
gates to the farm turnouts at 5f and 5h respectively.
Figure 3
A branching network irrigation scheme
HS 6 and HS7 are two sluice
September, 2009
Before computing the water level profile, these sluice

gates need to be calibrated.
With the data collected
6.10 m (channel bottom width), s=2 (channel side slope),

So=0.0016, and n=0.025.
of the calibrations are shown in Table 1.
about 732 m.
Gate
Flow regime
Gate status
Gate flow equation
HS3
NA
Closed
NA
Chows direct step computation.
Q = Cnf Ly
comparison.
Cnfw = Cnf / 2 g
Open
Submerged
Orifice Flow
Cso =0.3148 (yd /Go)

HS7
Submerged
Orifice Flow
close to what Chow computed.
Figure 4 is the plot of
the comparison of the computed water levels.
-0.2917
Q = Cso Lhd 2 g ( yu yd )
Open
RMSE (Root Mean Squared Error) of this
which indicates that the computation of this study is very
Q = Cso Lyd 2 g ( yu yd )
Open
Table 3 shows this
computation with Chows computation is 0.011219 (1%),
C nfw =0.0469
HS6
Chow used two methods for the computation: the

The computation of this study was compared with
1.5
u
Free Weir
Flow
The length of the profile is
graphical-integration method and the direct step method.
scheme.
HS4
23
This channel carries a discharge of Q=11.33 cms with b=
during the field survey and flow measurement the results
Table 1 Gate calibration in the branching network irrigation
Vol. 2 No.3
Table 3 Computed water level profiles of the trapezoid
Cso=0.609677(yd /Go) -2.2873
channel (1.53 m water depth behind the dam, Q=11.33 cms,
With all of the data collected in field survey and flow

measurement,
and
gate
calibration equations,
the
modeling tool computed the water level profile over the

branching network irrigation scheme automatically.
group of measured data was used to initialize model
b=6.10 m, s=2, So=0.0016, n=0.025, and length of the

profile is 732 m)
x distance to the
channel downstre
am end /m
y computed water y computed water

level profile in this
level profile by
study /m
Chow (1959)[6] /m
[(y-y)/y]2
0.00
1.52
1.52
47.09
1.54
1.53
3.92E-06
3.45E-05
97.22
1.56
1.55
148.86
1.59
1.57
9.32E-05
206.58
1.63
1.61
0.000129
270.38
1.67
1.65
0.000218
347.85
1.74
1.71
0.000256
396.46
1.79
1.76
0.000242
Table 2 A group of measured data for model computation
455.70
1.85
1.82
0.000225
initialization and verification.
493.68
1.90
1.87
0.000214
539.25
1.95
1.93
0.000159
computation (from main and lateral downstream ends)

and to verify the computation results at some check point
through the network scheme.
This group of data is
listed Table 2.
Data
Usage
575.70
2.00
1.97
0.000151
Head on HS5
0.22 m
Initial condition
622.79
2.06
2.04
7.97E-05
Gate Opening at HS4
0.30 m
Boundary condition
663.80
2.12
2.10
5.22E-05
Discharge over HS5
0.35 cms*
Initial condition
721.53
2.20
2.19
3.09E-05
RMSE
0.011219(1%)
Gate Opening at HS6
0.37 m
Discharge through HS6
0.01 cms
Boundary condition
Initial condition
Gate Opening at HS7
0.58 m
Boundary condition
Discharge through HS7
0.03 cms
Initial condition
Depth upstream of HS7
0.255 m
Initial condition
Depth upstream to point 5
0.67 m
Verification
Head on HS1
0.21 m
Verification
Note: *cms cubic meter per second.
Results and discussion
6.1
Single irrigation canal channel

Chow (1959) [6] gave an example of computing the
subcritical water level profile in a trapezoid channel.

This profile was created by a dam which backs up the
Figure 4 Plot of computed water level profiles of the trapezoid
water to a depth of 1.53 m immediately behind the dam.
channel
24
6.2
September, 2009
Vol. 2 No.3
4 compared with the measured data at check points. The
Branching network irrigation scheme

All of the computing results are summarized in Table
following figures illustrate the numbers in the table.
Table 4 Modeling tool computation result summary

Initial and computed
discharge /cm s-1
Location
Computed water depth

/m
Main downstream
0.35
Channel 6-7 (244 m)
0.35
0.49 m
Channel 5e-5g (15 m)
0.03
0.23 m
Channel 5b-5e (122 m)
0.01
0.37 m
Channel 4-5 (244 m)
0.4
Channel 2-3 (1,219 m)
0.4
1.07 m
Main upstream Head on HS1
0.4
0.2 m
Figure 5 shows the computed water level profile in

channel length is about 244 m.
0.22 m
Initial conditioned
0.26 m
Initial conditioned and the relative error between

computed and measured water depth is 11.5%
0.67 m
Based on the computation, the 0.67 m depth happened

at about 6.1 m upstream point 5
0.21 m
The relative error between computed and measured

water depth is 5%
length is about 122 m.
The computed water level is very
The
close to the normal depth at the distance of 91 m. So, if
The computed water
the distance is greater than 91 m, the flow can be
level is very close to the normal depth at the distance of

3.05 m.
Comment
Initial condition
0.67 m
the downstream channel of the main canal (6-7).
Measured water depth

/m
considered uniform at the depth of 0.37 m.
So, if the distance is greater than 3.05 m, the
flow can be considered uniform at the depth of 0.49 m.
Figure 6 Computed water level profile in the downstream

channel of the lateral 5e-5g.
Figure 5 Computed water level profile of the main canal
channel 6-7

the downstream channel of the lateral (5e-5g).
The
channel length is also about 15 m. The graphic indicates

that the water level is going to but never reaches the
normal depth.
So, the flow in this channel is considered
as pure gradually varied.
The average water depth is
about 0.23 m, which is 0.03 m away from the measured

depth of 0.26 m (relative error is 12%).
the upstream channel of the lateral (5b-5e). The channel
Figure 7 Computed water level profile in the upstream

channel of the lateral 5b-5e
September, 2009
Vol. 2 No.3
25
Finally, the computation produced 0.2 m water head
The channel
on the sharp-crested weir HS1, which is close to the
length is also about 244 m. The result indicates that the
measured depth of 0.21 m with the absolute error of 0.01
computed water level is going to but never arrives at the
m (relative error is 5%).
the intermediate channel of the main (4-5).
normal depth over the channel.

considered as pure gradually varied.
So, the flow is
Conclusions
Using the depth
curve, it can be derived that the 0.67 m water depth

happened about 6.1 m upstream point 5, which matches
the field measurement point.
This study has developed a tool for modeling the

water flow profile in irrigation distribution networks.
The modeling tool assumes SGVF flow in the branching
canal networks.
The tool starts the computation by
initializing discharge and water depth at the end of the

main canal and the laterals.
It handles the branching
networks in arbitrary layouts with first-level laterals.

The method can be extended to the cases of arbitrary
branching networks with second-level, third-level, and
n-level laterals.
The modeling tool was evaluated in
water flow profiling for a single irrigation canal channel

and an irrigation branching canal network.
The
calculations of the modeling tool had a 1% RMSE

compared to the benchmark calculation of a single
channel flow and 5% to 12% relative errors compared to
Figure 8 Computed water level profile in the intermediate
channel of the main 4-5
Figure 9 shows the computed water profile in the

upstream channel of the main (2-3). The channel length
is about 1219 m. The water level is close to the normal
depth at the distance of 244 m from the downstream end
of the channel.
So, if the distance is greater than 244 m,
check point measurement along a branching canal

network.
The implementation and results of the
modeling tool indicated a strong capability in handling

the modeling task in different conditions such that the
modeling process with the tool becomes simple, fast,
reliable and accurate with much less cost and least
configurations compared to commercially available
models and software packages.
The outcome of this
the flow can be considered uniform at the depth of
study will be able to play an important role in water
1.07 m.
quantification for planning, analysis and development for

modernization of irrigation systems for irrigation districts
in the Lower Rio Grande Valley of Texas and any other
similar areas.
Acknowledgements
The authors thank Mr. Milton Henry, Director of
Engineering & Technical Services, Jamaican National
Irrigation Commission for his work in field survey and
flow measurement.
This work was partially funded with
support from the Rio Grande Basin Initiative supported

by USDA-Cooperative State Research, Education, and
Figure 9 Computed water profile in the upstream channel
of the main 2-3
Extension Service (project no. 2005-45048-03208) and

administered by the Texas Water Resources Institute.
26
September, 2009

[5]
[References]
[1]
network. Journal of Irrigation and Drainage Engineering,

2008; 134(1): 4959.
for Water Resources, US Army Corps of Engineers, 1994.

Wallender W W. Furrow Irrigation Model Development and
[6]
[7]
Technical Completion Reports. Paper 658, 1989.
T.
Open-channel
hydraulics.
New
York:
Burt C M, Gartrell G. Irrigation canal simulation model usage.

119(4): 631636.
irrigation models. J Agric Engng Res, 2001; 79(4): 459

[4]
Journal of Irrigation and Drainage Engineering, 1993;
Esfandiari M, Maheshwari B L. Field evaluation of furrow

[8]
479.
Chow
McGraw-Hill Book Company, 1959.
Evaluation. University of California Water Resources Center.

[3]
Islam A, Raghuwanshi N S, Singh R. Development and

application of hydraulic simulation model for irrigation canal
Wurbs R A. Computer models for water resources planning

and management. Water Resources Support Center, Institute
[2]
Vol. 2 No.3
Huang Y. Water measurement for gate calibration. Biological
Hidalgo M M, Alandi P P, lvarez J F O, Martn-Benito J
and Agricultural Department, Texas A&M University.
M T. Calibration of on-demand irrigation network models.
Technical report, 2004.
Journal of Irrigation and Drainage Engineering, 2008;
[9]
USDI. Water measurement manual. 3rd Ed. U.S. Department

of the Interior, Bureau of Reclamation, 2001.
134(1): 3642.
IJABE has been covered by Inspec

The
International
Journal
of
Agricultural
&
Biological Engineering (IJABE, http://www.ijabe.org)
Appendix: Email notice from Team Leader of

INSPEC Acquisitions
launched in 2008 has been accepted for coverage within

the Inspec Database, after being successfully covered by
Dear Dr Yingkuan,
Chemical Abstracts(CA), CAB Abstracts databases, and

the CAB Full Text Repository.
Many thanks for your email, and apologies for the delay
The Inspec database is the world's largest and most
in replying.
comprehensive English-language bibliographic database,
I am pleased to confirm the International Journal of
in physics, electronics and computing. Developed from
Agricultural & Biological Engineering has been accepted
'Science Abstracts', first published in January 1898,
for coverage within the Inspec Database as from the
Inspec now scans over 4,000 journals (and other serial
issues that you have sent.
publications) worldwide, 2,000 conference proceedings
Please can you send issues as and when they are
and numerous books, reports, articles and dissertations,
published to myself at the address mentioned below.
for inclusion in the database. From 1969 to the present

day, the size of the database has grown to ten gigabytes
Kind regards,
of data and contains around ten million bibliographic

abstracts and indexing to journal articles, conference
Jason Foulsham
proceedings, technical reports and other literature in the
Team Leader, Acquisitions
fields of science and technology.
The Institution of Engineering and Technology
Moreover, IJABE has been listed in the world's
www.theiet.org
biggest open access journal platform, Open J-Gate
Michael Faraday House
(http://www.openj-gate.com) and Directory of Open
Six Hills Way
Access Journals (http://www.doaj.org), which is very
Stevenage SG1 2AY
helpful to increase the visibility of IJABE.
United Kingdom
(By Wang Yingkuan)
Vehicle System Dynamics
ISSN: 0042-3114 (Print) 1744-5159 (Online) Journal homepage: http://www.tandfonline.com/loi/nvsd20
Review of the State of Development of Advanced

Vehicle Control Systems (AVCS)
STEVEN E. SHLADOVER Deputy Director
To cite this article: STEVEN E. SHLADOVER Deputy Director (1995) Review of the State of
Development of Advanced Vehicle Control Systems (AVCS), Vehicle System Dynamics, 24:6-7,
551-595, DOI: 10.1080/00423119508969108
To link to this article: http://dx.doi.org/10.1080/00423119508969108
Published online: 24 Jun 2010.
Submit your article to this journal
Article views: 317
View related articles
Citing articles: 94 View citing articles
Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=nvsd20
Download by: [IIT Indian Institute of Technology - Mumbai]
Date: 15 June 2016, At: 23:37
Vehicle System Dynamics, 24 (1995), pp. 551-595
0042-31 14/95/2306-551$6.00
0Swets & Zeitlinger
Review of the State of Development of Advanced
Downloaded by [IIT Indian Institute of Technology - Mumbai] at 23:37 15 June 2016
Vehicle Control Systems (AVCS)

STEVEN E. SHLADOVER*
SUMMARY
This paper provides a broad review of the developments that have taken place within the past thirty
years in the field now known as Advanced Vehicle Control Systems (AVCS). This long time horizon
was chosen to ensure that significant findings from earlier generations of work are not overlooked
today. Following a history of the international development of AVCS, several methods of classifying
AVCS are introduced. The main body of the paper reviews the relevant literature in lateral, longitudinal and integrated control of road vehicles and summarizes the most significant findings from
this work.
I . DEFINITIONS O F ADVANCED VEHICLE CONTROL SYSTEMS
The term "Advanced Vehicle Control Systems" (AVCS) has been adopted
within the United States to refer to the subclass of Intelligent Transportation
Systems (ITS) that provide the driver with safety warnings or assistance in
controlling his vehicle, including the extension to full control of vehicle motions.
The term AVCS was defined by the ad-hoc working group that called itself
Mobility 2000, which worked during the period 1988-1991 to establish a U.S.
program for development of ITS. It originally established four functional groups
to define ITS technologies and services (which were known until recently as
Intelligent Vehicle Highway Systems or IVHS):
1.
2.
3.
4.
Advanced Traffic Management Systems (ATMS)

Advanced Driver Information Systems (ADIS)
Automated Vehicle Control Systems (AVCS)
Commercial Vehicles (CV)
These functions have since been somewhat redefined and augmented with
additional functional groups. In the original thinking, the distinction between
AVCS and ADIS was based primarily on the applicable time scale, with those
*Deputy Director, California PATH Program, University of California, Berkeley, Institute of
Transportation Studies, Richmond Field Station, 1357 S. 46th Street, Richmond, CA 94804,
U.S.A. Tel: 5 10-23 1-9537, Fax: 510-23 1-9565, e-mail ses@its.berkeIey.edu
552
STEVEN E. SHLADOVER
functions that involve decisions that must be made or actions that must be
implemented in less than 1 second being considered AVCS, while the slower
functions would be ATIS. The "A" in AVCS was later changed to be
"Advanced" in order to be directly comparable to ATMS and ADIS.
The words "Advanced vehicle control systems" are in fact a somewhat
misleading description of the technologies and services that comprise AVCS.
Many of the AVCS systems are not in fact control systems, but rather are systems
that provide warnings of hazardous conditions to drivers or enhance the ability
of drivers to perceive hazards themselves. The emphases in the designs of these
systems are on issues of sensing, sensor data processing and human factors rather
than on vehicle dynamics or control. In light of this, the most recent classification
of ITS user services developed for the U.S. National Program Plan for ITS has
replaced the term AVCS with the more comprehensive term, "Advanced Vehicle
Control and Safety Systems". While AVCS is excessively limiting terminology in
this context, it is also excessively broad in a different sense, because it is not meant
to include a variety of systems that could legitimately be considered to be
advanced vehicle control systems, such as traction control, antilock brakes,
active suspensions and four-wheel active steering systems. These can in some
sense be considered enabling technologies for AVCS but they are not generally
regarded as being AVCS in themselves.
2. HISTORY OF DEVELOPMENT OF AVCS

Although the term AVCS was coined in 1988, the history of development of these
systems begins much earlier. Interestingly, it really seems to begin with the most
advanced of the AVCS, the automated highway system (AHS), which was a
feature of the General Motors "Futurama" exhibit of future transportation
opportunities shown at the 1939-40 New York Worlds Fair.
During the late 1950s, General Motors and RCA developed and demonstrated (on test tracks) automatic control of steering and speed of automobiles
[I ,2] for what they called the "Electronic Highway." By 1962, the first university
research on automatic steering control of automobiles was reported from Ohio
State University (OSU) [3], under the sponsorship of the Ohio Department of
Highways and the U.S. Bureau of Public Roads. This led to a later, long-term
research program at OSU [4] on both steering and longitudinal (spacing) control,
under the sponsorship of the Ohio Department of Transportation and the
Federal Highway Administration (FHWA), from 1965 to 1980. The first
broad-scale investigation of the application of automation technologies to
urban transportation problems appears to have been in the M.I.T. Project
METRAN, in the spring of 1966 [5].
Major United States government involvement in the pursuit of roadway
automation began with the New Transportation Systems Research Act of 1966,
which instructed the federal Departments of Housing and Urban Development
STATE OF DEVELOPMENT OF AVCS
553
and Commerce to study new systems for urban transportation (since the
Department of Transportation did not yet exist). Thls "New Systems Study"
led to the creation of a series of reports from the Stanford Research Institute
(SRI) and General Research Corporation (GRC), culminating in the summary
report, "Tomorrow's Transportation: New Systems for the Urban Future" [6] in
1968. Six new urban transportation systems were recommended for development
in that report, one of which, called "Dual Mode," was essentially the automated
highway using vehicles that could operate in either a manual or an automated
mode (hence the name). The work on dual mode and automated highway
technologies was reported in a Transportation Research Board (TRB) conference in 1974, although the Proceedings [7] unfortunately did not contain full
texts of many of the papers.
In addition to the ongoing work at Ohio State, the FHWA sponsored broader
systems studies of the feasibility of an automated highway system, leading to
reports in 1969 [8] and 1977 [9]which recommended several areas needing further
research. A later FHWA study by General Motors in 1980-82 [lo, 111 examined
the technical potential of the automated highway in the context of realistic
implementation problems. All of these studies addressed broad operating
concepts, system-level tradeoffs and economic issues rather than detailed
engineering designs.
Privately-funded work on the development of roadway automation technologies is not generally reported because of proprietary restrictions on release of
information. This makes it difficult to tell how much activity there actually has
been in this area. Certainly General Motors has been active in the work on
roadway automation technology since the time of their original work, although
only limited glimpses of this have been available to the research community in
general [l2].
A large portion of the technology development work relevant to roadway
automation has actually been performed in pursuit of Automated Guideway
Transit (AGT) systems, formerly referred to as Personal Rapid Transit (PRT)
systems. This activity in the United States was primarily sponsored by the Urban
Mass Transportation Administration (UMTA) during most of the 1970s, and
produced a large body of research results which must be absorbed and understood by anyone hoping to advance the technologies for AVCS. The ACT
researchers had to address most of the issues that must be solved for AVCS,
except for the problems associated with transitions to and from automatic
control and with developing vehicles that can operate under both manual and
automatic control. The results o f t h e AGT research are scattered across an
extensive range of reports, journal papers and conferences. The central references, from which a more comprehensive literature search can be started, are the
four conference proceedings [I 3- 161.
The early developments in the United States were paralleled in Europe and
Japan. In England, the Road Research Laboratory conducted experiments on
automatic steering control in the late 1960s [17], while in Japan MITI's
5 54
STEVEN E. SHLADOVER
Mechanical Engineering Laboratory succeeded in controlling a car without a

driver at 100km/h by following an inductive cable in 1967 [18]. MITI's
Intelligent Vehicle Systems (IVS) program then demonstrated automatic steering control by use of binocular machine vision at 30 km/h in 1977 [19]. The
automotive applications were eclipsed by AGT activities, leading to significant
longitudinal control system developments in the Cabinentaxi system in Germany
[20], the CVS system in Japan [21], and the ARAMIS system in France [22], all
during the 1970s.
The "modern" history of AVCS development began in 1986 with parallel
actions in Europe and the U.S. In Europe, the PROMETHEUS Program
(PROgraM for European Traffic with Highest Efficiency and Unprecedented
Safety) was initiated by the motor vehicle industry, with funding support from
the European Community's EUREKA program and the governments of the
major western European countries. This program was intended to provide a
framework for cooperative development at the pre-competitive stage of technologies that could be used by the European motor vehicle industry and supplier
industries to improve the capacity and safety of the road transportation system
while also improving the world-wide competitive standing of the European
industries relative to their counterparts on other continents. Over its eight-year
span, this program invested about 650 MECU on a variety of projects that
ranged from basic research on sensor and chip fabrication methods and artificial
intelligence technologies to the development of a set of "Common European
Demonstrators", which were test vehicles equipped with systems that were
demonstrated to visitors driving in real traffic on the outskirts of Paris in
October 1994.
The European research programs have produced several system-level concepts that address nearer-term AVCS implementations such as "Cooperative
Driving" and the "Copilot". Cooperative Driving, as described by Broqua, et al
[23], includes four functions:
(a) Intelligent (i.e., Adaptive) Cruise Control to maintain desired longitudinal spacings between vehicles and provide warnings to the driver of
unsafe spacings;
(b) Intelligent Maneuvering and Control (IMC) to provide coordinated lane
changing and overtaking;
(c) Emergency Warning (EW) to provide localized short-range warnings
about emergency conditions; and
(d) Medium Range Preinformation (MRP) to provide the driver with
advance information about downstream problems such as crashes,
severe congestion, fog or ice.
Ref. [23] described the adaptive cruise control concept in detail, including plots
to show predicted performance of platoons of equipped vehicles and predictions
of the impacts on traffic conditions.
555
The "Copilot" concept is more of an information management and warning

system for the driver, gathering data from a comprehensive sensor suite and
processing it to provide recommendations to ensure safe driving in potentially
hazardous environments. The experimental implementation of this concept in
the "ProLab 2" test vehicle was described by Saint Blancard et a1 [24].
More or less independently of this European initiative, the California
Department of Transportation (Caltrans) decided in 1986 that it needed to
promote the development of new technologies to provide an economically viable
way to improve the productivity and capacity of California's freeway system so
that it would be able to meet the anticipated future travel needs of the state's
economy. This led Caltrans to join with the Institute of Transportation Studies
of the University of California at Berkeley to found the PATH Program
(originally called Program on .Advanced Technology for the Highway, and
later renamed as Partners for Advanced Transit and Highways). PATH'S
original charter had three primary thrusts, which were in clean propulsion
technology, highway automation and road vehicle navigation and guidance.
Recognizing that the deployment of operational systems would require a
national effort, rather than just the effort of one state, Caltrans and the PATH
university researchers stimulated their counterparts in other states to join forces
in the creation of Mobility 2000 to define a program of national scope [25,26].
This led to the creation of the national IVHS (now ITS) program in the United
States. The specter of the very well-funded and competitiveness-driven program
in Europe provided some of the stimulus needed to get political and funding
support in the U.S.
The current Japanese government ministry sponsored projects have each
developed their respective concept papers. This can be confusing to outsiders
because the competing ministries sponsor projects that appear to be covering
much of the same ground. The projects that are relevant to AVCS are under the
following ministries:
(a) Ministry of International Trade and Industry (MITI) has sponsored the
Personal Vehicle System (PVS) and Super-smart Vehicle System (SSVS)
projects, which have been aimed at fully automated driving. Taniguchi et
al [27] described the experimental results of the PVS development of a
prototype automated van that drove itself around a test track under
complete longitudinal and lateral control using machine vision and
ultrasonic sensors to detect its position relative to the lane. The SSVS
project, as described by Tsugawa et al [28], has only progressed to the
stage ofconcept study and evaluation, not yet leading to the development
and testing of vehicles. Reflecting this ministry's interest in the automotive industry, the emphasis here has been on autonomous vehicles.
(b) Ministry of Transport has developed the concept of the "Advanced Safety
Vehicle" (ASV), with an emphasis on collision avoidance and problems of
aging drivers, reflecting this ministry's interest in traffic safety.
'
556
STEVEN E. SHLADOVER
(c) Ministry of Construction, as part of its Advanced Road Transportation

System (ARTS) project, conceived the "Super Expressway for Automated Driving - 21" (SEAD 21) [19]. This represents a fully automated
highway concept, including automation of freight movements in exclusive
goods distribution tunnels that would be located deep beneath the streets
of major cities in Japan.
(d) Ministry of Construction, through its Public Works Research Institute,
has also proposed the "Road-Vehicle Cooperative Driving System"
described by Nakamura et a1 [29]. This is a multi-stage concept, beginning with road-vehicle communications for safety warnings and progressing from that to collision avoidance and eventually automated driving.
Reflecting this ministry's interest in infrastructure construction, the
emphasis is on infrastructure-based communication.
In the United States, development of the driver warning and assistance systems
of AVCS is being conducted largely by private companies in the motor vehicle
and supplier industries. The National Highway Traffic Safety Administration
(NHTSA) of the U.S. Department of Transportation is providing some
support for development of the underlying crash causation data bases and
the development of performance specifications [30]. The development of fully
automated driving systems in the United States was only being pursued by the
California PATH Program [31,32] during the period from 1986 to 1993.
However, more recently the Federal Highway Administration (FHWA) of
the U.S. Department of Transportation has begun an Automated Highway
System program [33]. This has included a set of Precursor Systems Analyses,
conducted during 1993-94, which have only started to produce some first
published findings [34].
It is extraordinarily difficult to develop a clear and comprehensive picture of
the current state of development of AVCS world-wide because a large proportion (quite likely, a substantial majority) of the work is being conducted in the
form of proprietary development work by private companies rather than
publicly-funded academic research. This means that most of the results (as
well as the methodologies) are not being documented in the open literature, but
are being kept secret in order to enhance competitive advantage. Based on the
outcomes of some of the public demonstrations of developmental systems, these
proprietary systems may even be more advanced than the state of the art
described in the published theoretical papers.
3. STRUCTURED APPROACH TO THINKING ABOUT AVCS

The field of AVCS is so broad that it must be subdivided into smaller units in
order to be studied productively. This subdivision can be pursued in several
useful directions, which will be outlined here:
services
hierarchical system architecture
a functions
a types of control.
a \ user
The user services breakdown has been used as the organizing principle for the
National Program Plan for Intelligent Transportation Systems in the United
States, under development by ITS America [35]. This breakdown is useful for
communicating how AVCS would be perceived and used by the general public,
rather than for addressing in-depth technical issues. The current listing of
relevant user services under consideration in the U.S. is:
Advanced Vehicle Control and Safety Systems ( A VCSS)
Longitudinal Collision Avoidance
Rear-End Crash Warning and Control
Autonomous Intelligent Cruise Control (AICC)
Cooperative Intelligent Cruise Control (CICC)
Head-On Crash Warning and Control
Passing Warning and Control
Backing Crash Warning
Lateral Collision Avoidance

Lane ChangeIBlind Spot Situation Displays, Collision Warning and Control
Road Departure Warning and Control
Intersection Collision Avoidance
Vision Enhancement for Crash Avoidance
Safety Readiness
Impaired Driver Warning and Control Override
Vehicle Condition Warning
In-Vehicle Infrastructure Condition Warning
Pre-Crash Restraint Deployment
Automated Highway System
Note that this listing includes primarily items that do not involve extensive
consideration of vehicle dynamics and control issues. More of the emphasis is on
the driver as the primary agent for control of the vehicle, with the AVCS systems
providing additional information or assistance. This listing also does not include
much of the concept of "cooperative driving," which has received much more
attention in Europe than in the U.S.
558
STEVEN E. SHLADOVER
A second way of subdividing AVCS is to consider the information architecture that could be used to implement the AVCS services. This would be
transparent to the transportation system users, but is critically important to
system developers. Varaiya has proposed a hierarchical framework for organizing the design of an automated highway system, which has also proven to be
extremely useful for structuring thought about all ITS system designs [36]. This
framework disaggregates system elements in terms of their information span,
frequency of update of data, spatial scope and function. In this scheme,
information is exchanged between adjacent layers or among peer entities at
the same layer, providing a way of simplifying system design and clarifying the
functions that must be implemented by each entity. The five layers of the
architecture are:
1. Network - entire urban or regional transportation network
2. Link - section of road of the order of 1 km in length
3. Coordination - cluster of neighboring vehicles that must coordinate
motions to ensure safety
4. Regulation - closed-loop control implementation on an individual vehicle
5. Physical - sensors and actuators on individual vehicle.
The issues of vehicle dynamics and control are normally confined to the
regulation and physical layers within this architecture. The large majority of
the existing literature addresses these layers, since the higher layers are only
relevant to the most sophisticated and long-term system designs. However, the
literature on the higher layers is starting to grow vigorously.
Within the regulation and physical layers, the functions that must be
performed by an AVCS system can be subdivided using traditional block
diagram schematics of feedback control systems. The fully automated systems
are somewhat different from the warning and assistance systems, but the overall
schematics remain quite similar. Figure 1 illustrates the general functional block
diagram for AVCS warning and assistance systems, while Figure 2 shows the
analogous diagram for fully-automated AVCS systems. Each diagram includes
Raw Sensor
Sensor
Dynamics
Driver Actions
(Human Factors)
Driver
Interface
Fig. I(a). Functional block diagram for driver warning and assistance systems.
Sensors
Sensor
homing
Corn utotion
of L z a r ~
Warning
Driver
Interface
Human
Factors
Vehicle
Dynamics
Fig. I(b). Relative levels of technical challenges for driver warning and assistance systems.
an illustration of the author's personal evaluation of the relative level of

technical challenges associated with the development of each of the functions.
These should be considered analogous to estimates of the distribution of effort
and investment that will be needed to develop and implement AVCS services.
Note that vehicle dynamics and control issues are not likely to be as prominent as
issues associated with sensors and the software needed to interpret the sensor
signals. For a much more detailed functional breakdown of fully-automated
AHS, see Zhang et al. [37].
Since the IAVSD is focused much more on issues of vehicle dynamics and
control than on sensor or signal processing issues, the most useful organizing
principle for this review paper appears to be based on the types of control. Under
this approach, the AVCS control issues can be subdivided into:
lateral control
longitudinal control
integrated lateral/longitudinal control
higher-level control issues.
link or
Coordination
Raw Sensor
Data
Desired
Path
Sensor
Control
Control
Command
Actual
Trajectory
Actuation
Fig. 2(a). Functional block diagram for automated systems.
560
STEVEN E. SHLADOVER
Senran
Sensor
Processing
Control
Computation
Vehicle
Dynamics
Control
Actuation
Fig. 2(b). Relative levels of technical challenges for automated systems.
4. LIMITATIONS TO THE LITERATURE REVIEW

The review of the AVCS literature that follows is subject to several limitations,
which need to be highlighted at the start:
The existing literature on AVCS is vast, because of the breadth of the subject
and the substantial time that has elapsed since serious research on AVCS began.
It is impossible to give a truly comprehensive review in a paper of less than book
length, so attention will have to be focused on highlights. Given that the author's
native language is English, the concentration is clearly on English-language
literature, so the literature in other languages has been largely neglected.
The literature is seriously fragmented because of the broadly interdisciplinary
nature of AVCS. Each technical community has its own favorite conferences and
journals, which means that it is necessary to refer to a great many sources in
order to obtain a comprehensive view of AVCS. Examples of some of the key
sources for different types of information are:
General system descr@tions and developmenf sfatus:Transportation Research
Board (TRB) Annual Meeting, IVHSIITS America Annual Meeting, Transportation Research Records, ITS World Congress, ITS Review, SAE Future
Transportation Technology Conference, IEEE Vehicle Navigation and Information Systems Conference (VNIS), IVHS Journal;
Historical developments (1960-1980 time frame): Personal Rapid Transit
(PRT) Conference proceedings, Transportation Research Records, IEEE
Transactions on Vehicular Technology, ASME Journal of Dynamic Systems,
Measurement and Control;
Vehicle Dynamics and Control issues: ASME Journal of Dynamic Systems,
Measurement and Control, Vehicle System Dynamics, IVHS Journal, American
Control Conference Proceedings, IEEE Conference on Decision and Control
Proceedings, AVEC Symposium Proceedings;
Human Factors and Safety issues: Accident Analysis and Prevention, Transportation Research Board Annual Meeting, ITS America Annual Meeting,
Human Factors Society meeting;
56 1
Machine Vision and Autonomous Vehicle Technology: IEEE Intelligent

Vehicles Symposium, Association for Unmanned Vehicle Systems (AUVS)
Annual Meeting; Society of Photo-Optical Instrumentation Engineers (SPIE)
conferences;
Higher-level control issues: IEEE Conference on Decision and Control, IEEE
Transactions on Automatic Control, IVHS Journal, Transportation Research
Part C.
There is a serious paucity of real vehicle test data reported in the literature.
Most papers have been based on analyses and computer simulations, without
experimental verification. This is in large part because of the serious investment
of time and resources needed toimplement AVCS experimentally, but nevertheless it significantly limits the usefulness of much of the literature. Even the
papers that include some experimental results tend to be rather vague about
quantitative values, and rely more heavily on plots of test results that give the
reader only a qualitative sense of what has been accomplished. This makes it
difficult to assess the true value of the reported results or to compare the results
achieved by different researchers. Because of proprietary concerns, most of the
existing experimental data have not even been reported publicly.
Much of the literature dating from the "modern" revival of interest in AVCS
has been written in ignorance of the accomplishments (and failings) of the 1960s
and 1970s.As a consequence, there has been some inefficiency in the research and
the recent literature has not been as strong as one might have hoped. The longhonored research practice of conducting a thorough literature search at the start
of a new undertaking appears to have fallen into some disuse.
5. REVIEW OF LATERAL CONTROL RESEARCH

ACCOMPLISHMENTS
5.1. Earliest Activities
Electronically guided steering for scale-model automobiles was reported as early
as 1953, and for full-scale automobiles in 1957 [38]. The first published reports of
an attempt to automate the steering control of an automobile were made by
General Motors and RCA in the late 1950s, following a classic series of
experiments in Princeton, NJ [l, 39-41]. The steering reference used by GM
and RCA was a wire installed along the centerline of a roadway lane carrying a 2
KHz electric current, which was sensed by inductive coils mounted on the
vehicle. The steering control law design has not been documented, but was
apparently an empirical attempt to demonstrate the concept, using analog
control circuitry to provide inputs to an electrohydraulic steering servo.
Analogous efforts followed during the 1960s at the Road Research Laboratory (RRL) in England, Ohio State University in the U.S. and MITI's Mechanical Engineering Laboratory in Japan. The RRL implemented automatic steering
control on two different vehicles, a DS Citroen using an electrohydraulic servo
562
STEVEN E. SHLADOVER
based on the vehicle's own power steering system [17] and a Ford Cortina using
an electromechanical servo system [42]. Both steering servo actuation systems
proved to be problematic in this research. RRL made a serious commitment to
automatic lateral control research, as evidenced by their installation of 16 km of
reference cable in the outer lane of the M4 motorway near Reading, England [43].
5.2. Ohio State University Research

Ohio State University's (OSU) research on automatic lateral control during the
1960s and 1970s, under the leadership of Electrical Engineering Professor
Robert Fenton, is the most thoroughly documented of any conducted during
that period, having been reported in dozens of technical papers and reports. See
[4] for a very useful review of that entire research program, including more than
forty reference citations. The OSU research employed classical control design
methods to develop linear (and later on, nonlinear) steering compensators for
their test vehicle. They appear to have published the first detailed theoretical
study of automatic steering control [3], with comparisons of the frequency and
transient response characteristics of four different controllers for application to
an overly simplified single degree of freedom vehicle model. They experimented
with both single and double reference wires in the pavement, eventually settling
on the single wire system and using the phase characteristics of the sensor signals
rather than amplitude because of better immunity to magnetic field distortions
caused by reinforcing steel in the pavement [44-461.
The OSU group conducted experiments at speeds up to 36m/s on an
unopened section of highway using 1965 and 1969 model Plymouth automobiles. They were able to report automatic steering with a tracking error in a gentle
curve (0.1 g lateral acceleration) that was generally within 2.5 cm and went up to
3.8cm in a wind gust [47]. This controller was based on classical frequency
domain design for a linear two-degree-of-freedom vehicle dynamic model with
roll-steer coupling effects included.
Later OSU research on lateral control departed from the active wire reference
approach because the sponsoring agency, FHWA, decided that active roadway
references were not suitable for operations and maintenance considerations [4].
Fenton and Selim [48] adopted an optimal linear control design approach using
an observer to develop full state feedback from a measurement of lateral offset at
the front of the vehicle. They made their controller velocity adaptive to account
for the velocity dependence of vehicle lateral dynamics by doing separate optimal
designs for each 1 m/s velocity step between 10 and 35 m/s. This design was tested
using an analog computer simulation of the vehicle.
5.3. Other Empirical Research of 1960s and 1970s
The approach adopted by Dr. Sadayuki Tsugawa of MITI's Mechanical
Engineering Laboratory in Japan and his colleagues during the 1960s and
1970s began with wire-follower systems, but then shifted to the use of machine
vision technology, with video cameras mounted on the test vehicle, and image
563
processing software used to identify the location and alignment of the vehicle
relative to the lane markers [49,50]. This work was derived directly from
technology developed for factory automation, where speeds are low and vehicle
dynamic effects are not strong.
All of the approaches mentioned thus far were based on use of passive sensors
mounted on the automatically steered vehicles. An additional concept, developed by Spangler and Snell[51] and elucidated by TRW [8], was based on the use
of a radio frequency transmitter on the vehicle inducing currents in a passive
conductor beneath the road surface, and inductive coils on the vehicle then
sensing the magnetic field produced by these currents. This system was of course
very sensitive to cross-talk between the transmitter and the sensor coils.
A variety of automatically steered vehicle applications were reported in the
1960s and early 1970s, but none of them seemed to endure for long. These
included an electronic steering control system developed for a full-size passenger
bus by Barrett Electronics in 1961 [52], and vehicles that operated at very low
speeds in hospitals and warehouses [53]. The sensing aspect of lateral guidance
was used to provide position error information to human drivers steering
military vehicles through Arctic wastelands in adverse weather [54].
Much research was done in the 1960s and 1970s on mechanical guidance of
automatically steered vehicles, for application to automated guideway systems
and busways. Various of these systems have been implemented and are operating
in public service today, with performance that is in some cases excellent. This
work is not addressed here because its dependence on the availability of a special
mechanical reference is not compatible with AVCS needs to operate in more
general roadway conditions.
5.4. Developing Analytical Foundations for Lateral Control

With the exception of [3], the lateral control developments of the 1960s and early
1970s were not based on strong analytical foundations, but rather were much
more empirically based. This began to change with the work of Pasternack
[55,56], who first applied modern control principles to the analysis of automatic
lateral control, in order to determine the controllability of a simple two degree of
freedom vehicle lateral dynamic model using a single measurement of lane
tracking error obtained from a sensor mounted forward of the vehicle center of
mass. This work was followed by the parallel studies by Bonderson at General
Motors [57] and Shladover at M.I.T. [58], applying optimal control theory to
design optimal lateral controllers. Bonderson applied time-domain optimal
linear regulator theory (solving the matrix Riccati equation), while Shladover
applied a frequency domain version of Wiener filter theory to the lateral control
problem, producing comparable results. The work by Shladover included the
definition of a fundamental trade-off between ride quality and tracking accuracy
which is identical to the ride qualityltracking accuracy trade-off that was
previously identified for suspension systems of high-speed ground transportation vehicles.
564
STEVEN E. SHLADOVER
The M.I.T. lateral control research continued with use of optimal linear
regulator theory for lane following [59], and also addressed transient effects
associated with curve entry [60]. A very similar technical approach was adopted
by Panik and his colleagues at Daimler Benz [61], with the important addition of
an experimental verification of the control system design on a full-scale
passenger bus, using an electronic wire guidance reference [62,63]. This work
is one of the most important contributions to lateral control because it has been
extended to operational tests in actual passenger service in Fiirth, Germany
(electronic reference), as well as being used in daily transit operations in Essen,
Germany and Adelaide, Australia (mechanical reference). The electronic
reference system has more recently been applied for lateral control of the
special-purpose vehicles that operate in the service tunnel of the Eurotunnel
(the English Channel tunnel).
5.5. Applications of Machine Vision Technology

All of the work described thus far has been conducted by researchers from the
electrical and mechanical engineering communities, making use of special
references installed in or on the roadway. A largely separate line of research
has been conducted by the artificial ir~telligencecommunity, making use of
machine vision technology to recognize the orientation of the vehicle relative to
the conventional roadway markings. Typically, these researchers have focused
their attention on issues of image processing rather than on those of vehicle
dynamics and control. This work is descended from the defense industry's
interest in guidance of autonomous off-road vehicles, and the underlying
research has generally been funded by defense agencies. The most coherent
justification for application of this technology to road vehicles is in the paper by
Metzler of Daimler Benz [64].
The pioneering work in road vehicle guidance using machine vision has been
done by Prof. Ernst Dickmanns of the Universitat der Bundeswehr in Munich
and his colleagues [65-671. The literature on lateral guidance using machine
vision is extremely large, but most of it is unlikely to be of much interest to the
vehicle dynamics and control community because it rarely if ever addresses
vehicle dynamics or control issues. Some of the more interesting examples that
use vision for both lateral and longitudinal control will be covered later, in
section 7 on integrated lateral and longitudinal control. For recent reviews of the
state of the art in this application of machine vision technology, it would be best
to refer to the book and conferences organized by Masaki [68-701.
5.6. PATH Program and Recent Related Lateral Control Research

Automatic lateral control has been researched by the California PATH Program
at the University of California, Berkeley since 1988, primarily based on the
concept of vehicles sensing their position relative to a series of permanent
magnets embedded in the pavement along the lane centerline. This work
began with development of vehicle dynamic models and several control laws,
565
which were evaluated using simulations before being tested on vehicles. The
analytical basis for this work, including the feedforward of roadway curvature
information, is described by Peng and Tomizuka [71,72]. They used a frequencyshaped linear quadratic (FSLQ) optimal controller to explicitly address the
maximization of ride quality, while the roadway geometry preview helped keep
the feedback gains low. Peng et al. [73,74] conducted experiments in cooperation
with IMRA America to demonstrate the performance of this controller and to
compare it with a simpler PID with preview controller. These experiments
showed excellent ride quality and curve tracking accuracy, including fairly
aggressive curving (0.27 g lateral acceleration) and anomalous operating conditions (extra payload weight, partially deflated tires in different positions, snowy
road surface). The lateral tracking errors remained less than 10 cm under almost
all conditions, and less than 15cm in the worst cases. Very significantly, the
performance was shown to be measurably better than the steering of the same
test vehicle by a variety of human drivers [73], and the control system initially
designed for a Toyota Celica was converted for use on a Pontiac 6000 with a
minimum of modifications, while retaining excellent performance [74].
The PATH lateral control work has been extended to testing of an alternative
controller, using fuzzy control [75]. Controller design has been initiated for lane
changing maneuvers [76], using modified FSLQ and sliding controllers, but this
has not yet been proven experimentally. Two other lateral control design efforts
within PATH have been based on the idea of vehicles sensing the movements of
the vehicles in front of them, rather than sensing their location relative to the
roadway lane center (in order to save the cost of any special roadway
infrastructure). Narendren and Hedrick proposed "auto-tracking" using
vehicle-to vehicle range and azimuth angle measurements and a sliding control
law [77]. Fujioka proposed a trajectory-following controller, also using sliding
control, in which the preceding vehicle would need to communicate its tracking
error to the following vehicle [78]. Neither of these approaches is applicable to
the first vehicle in sequence (which would have no other vehicle to follow) and
neither has yet been verified experimentally.
Several researchers have addressed the robustness of lateral control systems
from different perspectives. Ackermann (79-811 has emphasized analytical
approaches to deriving fixed-gain controllers that can steer successfully under
a full range of speed, payload and road surface conditions and has shown
analysis and simulation results with good performance in the absence of preview
information about road geometry [80,81]. Patwardhan and Tomizuka [82] have
designed a control system that can be reconfigured to accommodate the change
in vehicle dynamics produced by a sudden tire blow-out. They have tested their
open-loop vehicle model against experimental results using a full-scale vehicle
with a sudden tire deflation, but have not yet tested the closed-loop controller.
Smith and Starkey [83] studied automatic lateral control for emergency
maneuvers that would require higher lateral accelerations than contemplated
in the designs of other lateral control systems. They showed howgain scheduling
566
STEVEN E. SHLADOVER
could be used to accommodate the tire nonlinearities that would be expected

under these extreme operating conditions (only in simulation, not yet experimentally).
5.7. Nearer-Term Lateral Guidance Applications

All of the research described thus far has been aimed at fully automatic lateral
control of vehicles. However, the nearer-term implementations of lateral
guidance and control technology are likely to be in the form of warning and
control assistance systems, consistent with the user services that were defined in
Section 3. The literature on these applications is not nearly so well developed,
although there are a few applicable papers. Kushiro et al[84] described a crosswind compensation system that relied on a cross-wind sensor to provide
feedforward information that could be used to provide a command to the
vehicle steering system. This was tested using vehicles equipped with steering
actuators on the front wheels only as well as vehicles with both front and rear
active steering control, and in both cases showed extremely effective reduction of
the crosswind's disturbance to the vehicle motion. Adachi et al. [85] described a
system for lane departure warning and lane departure prevention control using
video image processing, as part of the Japanese Advanced Safety Vehicle project.
Lemarchand and Coffin [86] described the initial human factors experiments
with a lane departure warning system that uses machine vision sensors and
steering wheel torque feedback to the driver as the control action. They tested 15
drivers on 48 km of driving, and found that the system needed to be able to adapt
to provide different performance on wide versus narrow roads.
5.8. Lateral Reference and Sensor Systems

The large majority of the lateral control research has been based on either of two
kinds of roadway center references - the active wire in the center of the lane or
the painted lines delineating the lane boundaries. Several other kinds of
reference/sensor combinations have also been studied. Mayhan and Bishel [87]
experimented with a short-range two-frequency radar that would bounce off a
railing installed alongside the roadway, providing about twice the tracking error
that their OSU colleagues had achieved with the wire follower system. The radar
for lateral control has been pursued more recently by Giubbolini and Cugiani in
Italy [88,89] and by Johnson, et a1 [90] in the U.S., who wanted to use radarreflecting marking materials on the road surface. Bishel has recently attempted
to use the railing-reference radar system as the reference for a test track where
heavy-duty trucks are steered automatically along very rough pavement for
endurance testing [91]. Johnston et al. [92] used discrete markers as the steering
reference for the very low speed operations of the "Automated Mixed Traffic
Vehicle". Although most of their discussion was about the general concept of use
of discrete markers, their paper described an optical system using retro-reflective
567
markers. Yasui and Margolis [93] described the design of a system that would use
retroflective markers between the lanes as the reference for an optical scanning
sensor, providing both preview and current vehicle orientation information to a
simulated lateral control system. Zhang and Parsons [94] proposed the use of
discrete permanent magnets as the reference markers for lateral control, and
conducted a series of experiments to prove that these could provide extremely
accurate (5 mm) reference position information to a vehicle equipped with
magnetometer sensors. These magnetic markers, which can be installed in a
sequence that incorporates a binary code (N or S poles up) to represent
upcoming road geometry changes, have served as the reference for most of the
PATH lateral control experimental research.
5.9. Summary of Lateral Control State of Knowledge
Based on the research conducted thus far, the current state of knowledge about
automatic lateral control appears to include the following:
A variety of referencelsensor combinations appear to be technically

feasible and have been demonstrated in operation on test vehicles.
These have differences in accuracy and susceptibility to environmental
disturbances, which will need to be weighed along with the differences in
the costs they impose on the vehicles and roadway systems. Ultimately,
the decision about what type of system to implement will be driven by the
economics.
Automatic lateral control systems have been demonstrated to have lane
tracking accuracy and ride quality that exceed the capabilities of normal
human drivers.
Providing preview information about road geometry changes enables the
steering control systems to significantly improve their ride qualityltracking
accuracy trade-off frontier.
Because there is no inherently safe degraded mode of operation for an
automatic steering system, it is critically important that lateral control
systems be provided with fail-safe capabilities.
It appears to be possible in theory for lateral control systems to adapt their
performance to abnormal operating conditions such as vehicle loading,
road surface friction, tire anomalies, etc., but this has not yet been proven
conclusively in vehicle experiments.
Robustness of lateral control system performance to a full range of external
forces has not yet been proven. This needs to include not only ambient wind
forces, but also the aerodynamic forces from other adjacent vehicles, tire
forces from road debris and surface anomalies, and moderate collisions
with other vehicles.
High-performance steering actuators (with small deadbands and lags) are
needed in order to get good lateral control system performance.
568
STEVEN E. SHLADOVER
6. REVIEW O F LONGITUDINAL CONTROL RESEARCH

ACCOMPLISHMENTS
Vehicle longitudinal control has a longer and richer history relevant to AVCS
than does lateral control. This is largely because methods of longitudinal control
have been applied to railroad and fixed guideway vehicle systems that were
steered by special mechanical means. Those lateral control (steering) technologies are not relevant to the lateral control of road vehicles in general, but the
longitudinal control methods are relevant.
While lateral control is a function of each individual vehicle relative to the
roadway reference, longitudinal control adds the complication of potential
interactions among multiple controlled vehicles. Much of the diversity in
approaches to longitudinal control is related to different ways of handling
these interactions or avoiding them. There are three fundamentally different
approaches to longitudinal vehicle control:
I. Fixed block control
2. Point following (or moving block control)
3. Vehicle following.
Fixed block control, which is the standard used in most railroad and automated
guideway systems, is the simplest and oldest approach, but the one with by far the
lowest vehicle capacity. In this form of control, the roadway is divided into
segments, called blocks, and vehicles are prohibited from entering any block that
is already occupied by another vehicle. This can be refined further in "multiple
aspect" block control, which permits much shorter blocks and establishes
intermediate criteria for operations by vehicles located different numbers of
block lengths behind an occupied block (thereby enabling vehicles to operate
closer together).
Point following systems establish moving virtual reference markers along the
roadside, which the vehicles are commanded to follow as accurately as they can.
This enables higher capacity than the fixed block systems, but with simpler invehicle control systems than the vehicle following systems. It requires installation of the roadside equipment to generate the reference markers and also
requires a certain degree of centralization of the system management functions
(scheduling of vehicle movements, with synchronous or quasi-synchronous
network management).
Both the fixed block and point following systems provide a basis for
controlling normal vehicle operations on unobstructed roadways. Failures of
vehicles are accommodated by roadside control actions (prohibiting vehicle
entry into a fixed block, or decelerating the reference marker to a stop for point
following). However, the vehicles are essentially "blind" to what is happening in
front of them unless they are provided with separate obstacle detection systems.
Such systems are likely to be needed to detect foreign objects that might intrude
in the roadway and could cause damaging crashes.
569
Vehicle following systems provide essentially all of the regulation-layer

control functionality on the vehicles, with none required to be on the roadside.
This shifts some of the system costs from roadway to vehicles, considerably
increasing the vehicle system complexity, but also provides the potential for
significantly higher lane capacity than the point following systems. The vehicles
must use ranging sensors to detect their state relative to those of their neighbors,
and these same sensors can also provide the obstacle detection function to avoid
crashes with foreign objects as well as with other vehicles. Vehicle following
systems are also compatible with asynchronous network management and a
more distributed (as opposed to centralized) system control philosophy.
Virtually all of the longitudinal control literature falls into one of the three
general categories defined above. However, there was one attempt by K.C. Chu
[95] to bridge the gaps between the categories by addressing the more basic issue
of the structure of the information available about the controlled vehicle and its
errors relative to another vehicle. Chu applied "optimal decentralized regulator
theory" to determine feedback coefficients suitable for each information
structure and then compared the performance of these separate optimum
solutions to evaluate their respective potential capabilities.
The review of relevant literature that follows proceeds through the three
different longitudinal control approaches in sequence, beginning with the
simplest proposals under each general approach and then proceeding to the
more complicated ones. Attention is confined to papers dealing with rubber-tired
vehlcles in order to concentrate on the work relevant to AVCS and to avoid the
vast literature on railroad control systems.
6.1. Review of Fixed-Block Longitudinal Control Systems
Conventional fixed-block systems make use of only binary vehicle location
information (either a block is occupied by a vehicle or it is not). This produces
very large quantization errors in determining the location of any vehicle, which
severely limits how closely the vehicles can be spaced on the roadway. A
discrepancy in vehicle location can only be detected after at least one block
length has been traversed (a vehicle has failed to appear where it was expected).
Application of this approach to automated guideway transit (AGT) systems
using rubber-tired vehicles was considered in [96,97], but these shortcomings
were identified and evaluated there.
Some of these shortcomings could be reduced by augmenting the position
information with velocity information as well, producing a concept called
"Augmented Block Guidance" by the Applied Physics Laboratory (APL) of
Johns Hopkins University [98-1001. APL considered a six-aspect block system,
in which a vehicle would be unconstrained when at least six blocks behind its
predecessor but would encounter increasing performance constraints as it
approached closer to the predecessor. Wheel rotation measurements would
permit further refinements to the estimate of the vehicle's position within its
block, while the vehicle's own speed measurement would also be factored into
570
STEVEN E. SHLADOVER
the control system's determination of a suitable response. Collision avoidance

was proposed by use of a phase-plane approach to decide when to activate
emergency braking (based on velocity and spacing to the failed vehicle location).
Simulations of response showed generally good performance, but with some
upstream propagation of disturbances.
Whitney and Tomizuka [I011 designed a PID controller to augment the simple
block system position information with rate (velocity) feedback and to remove
steady-state position errors. They showed it to be equivalent to the optimal point
following controller of Wilkie (to be described below), and effective for normal
operations. However, emergency conditions caused oscillatory performance and
increased stopping distances, which wouM have required the addition of
acceleration feedback to improve performance.
6.2. Review of Point-Following Systems
Point-following longitudinal control systems received considerable attention for
application to automated guideway transit (AGT) and personal rapid transit
(PRT) systems in the 1960s and 1970s, and some of these were implemented in
operational systems.
Two attempts were made to design point-follower controllers using only
longitudinal position error information (information about the distance between
the vehicle and its desired reference point). The first, by TRW, neglected the
important factor of vehicle longitudinal dynamics [8]. The second, by Smith
[102], applied optimal regulator control to the vehicles and used a Luenberger
discrete-time observer to estimate the remaining vehicle states from the longitudinal position error. This approach increased controller complexity to save on
referencelsensor system cost and complexity.
Several researchers explicitly added feedback of the vehicle's own speed to the
position error. Hinman [103] proposed feedback of vehicle speed and PI
compensation of the position error to achieve good dynamic response. For the
CVS system in Japan, Ishii et al. [21] applied PD compensation to the spacing
error in order to derive the velocity error estimate. CVS incorporated this in
something called the "two target tracking" scheme, which appears to be a form
of model following control that does not seem to have been documented in any
papers published in English. Model following control was also proposed using
this level of feedback information by Garrard [104].
Fenton and Chu [lo51 developed a point-following controller using the
spacing between the vehicle and its reference point and the difference between
the vehicle's speed and the target reference point speed as the measured variables.
They refined Fenton's earlier vehicle following controller [4] using PI compensation by including the nonlinear velocity dependence of tire slip and propulsion
system dynamics. Their controller sought to compensate for some of the
undesired nonlinearities by means of an approximate nonlinear pole-zero
cancellation [105]. Vehicle experiments at highway speeds for a simulated
57 1
merging maneuver produced maximum position errors of 42cm and final

position errors of 12cm, while a hard braking command (0.45 g) produced a
final position error of 1.2 m [105].
The most substantial concentration of effort on point-follower control has
been based on the assumption of feeding back complete information about the
state of the vehicle relative to its pointer, using optimal linear regulator theory.
The classic paper in this category was by Wilkie [106], who minimized a
quadratic sum of vehicle jerk and position, velocity and acceleration errors,
while assuming continuous availability of sensor data. Fling and Olson [107]
extended this to assume sampling of information from discrete roadside
markers. Further extensions were proposed by Garrard and Kornhauser
[108,109], including the )use of a higher-order model to address passenger
comfort concerns and state observers to estimate all the needed state variables
from measurements of position and velocity errors. Working separately,
Kornhauser [I 101 applied a Kalman filter to account for the effects of input
noise and finite sampling rates and Garrard [ I l l ] addressed the robustness of the
controller to external forces and changes in vehicle loading.
Only one research team, at the Alden Self Transit Systems Corporation,
appears to have addressed the issue of point-follower operation under abnormal
conditions, in which the decision must be made whether to apply emergency
braking [112-1141. In this study, the binary decision (apply brakes or not) was to
be based on position error exceeding a threshold value [I121 and then later
making that threshold a function of speed [I131 and acceleration [114].
The literature on point-follower longitudinal control appears to have ended
by 1980, which is about when it became apparent that the cost of providing the
higher level of in-vehicle intelligence needed by vehicle follower systems should
not be a deterrent to their use because the computing capabilities were becoming
so cheap with the widespread availability of microcomputers.
6.3. Review of Vehicle-Following Systems

The largest and richest body of research has been in vehicle-following longitudinal control systems, especially in more recent years. These systems are
inherently more complicated than the earlier ones, they offer more diversity in
possible system designs, and most importantly they have the potential to provide
substantially greater lane capacity. The great diversity in vehlcle-follower
concepts can be extremely confusing. In an attempt to impose some structure
on this heterogeneous topic, Figure 3 identifies the different types of information
that can be supplied to a controller located on the controlled vehicle (highlighted
in the box). Each change in information structure can have significant ramifications for system performance. Note that only the few shaded examples can be
implemented on autonomous vehicles (using sensor measurements without
communication). We will begin with the simplest controller structure, and
proceed from there to the most complex.
The earlier papers in each category typically were based on application of
572
STEVEN E. SHLADOVER
CONTROL STRUCTURE NUMBER
ROADSIDE
ommanded speed
LEADING VEHICLE
PRECEDING VEHICLE m
pacing to preceding vehicle m + 1
PRECEDING VEHICLE 1
acing to preceding vehicle 2
ommanded speed
TRAILING VEHICLE
pacing to controlled vehicle
peed
cceleration
TRAILING VEHICLE n
pacing to trailing vehicle n 1

peed
cceleration
Fig. 3. Sources of feedback information for vehicle-follower longitudinal control (variables available by autonomous sensing listed by bold face).
linear, time-invariant control system design approaches. By the late 1970s it had
become apparent that these were not adequate for addressing the important
performance issues, and from that time forward most relevant work has included
nonlinear, time varying and adaptive control considerations. Very few of these
structures have been implemented and verified experimentally on full-scale
vehicles (structures 2,3,4,8,9).
One of the most persistent and controversial issues in vehicle-following
control systems is the selection of the desired following distance. The three
basic classes of vehicle following are:
(a) constant spacing

(b) constant headway (time)
(c) constant safety factor.
The constant-spacing systems are most suitable for use at very short spacings,
particularly within close-formation platoons. When they are well designed, they
can provide the vehicle occupants with the illusion of a rigid mechanical coupling
between vehicles; however, they require more information than the other classes
in order to achieve asymptotic string stability. The constant-headway systems
can be made asymptotically stable using information derived only from sensors
on the following vehicle, but their spacing varies proportionally with operating
speed in order to maintain headway constant. The constant safety factor systems
may use any of several safety criteria, but typically they maintain spacing
proportional to v2 ("constant k-factor" vehicle following).
6.3.1. Feedback of Spacing and Vehicle's Own Speed (Structure I )
The simplest vehicle-follower control structure was studied by Brown [115], using
only measurements of the spacing to the preceding vehicle and the follower's own
speed. Brown's study showed satisfactory performance with P-I compensation
applied to the spacing error, combined with proportional feedback of the
follower's speed. Thls approach was not tested on vehicles or subjected to
evaluation with noise or external disturbances.
6.3.2.Feedback of Spacing and Velocity Difference Relative to Preceding Vehicle
(Structure 2 )
This is a very popular controller structure because of its simplicity and its
potential for use in mixed traffic, where only one or a few vehicles are equipped
for automatic longitudinal control. The velocity difference could be measured
directly by a range rate sensor on the controlled vehicle, or it could be derived
from the difference between the measurement of the controlled vehicle's speed
and the predecessor's own speed, communicated to the controlled vehicle. The
availability of the speed difference virtually guarantees separate knowledge of
the speeds of both vehicles, since every vehicle is equipped with a speedometer.
The first published technical papers describing this controller structure
appear to be by Fenton's group at Ohio State University [116-1181. They
divided the spacing-velocity difference phase plane into six regions and defined
different control laws for those regions. This controller design was simulated
extensively for the two-vehicle case and has also been tested on a full-scale
automobile on a test track using a "phantom" lead car or a mechanical take-up
reel to provide the spacing and velocity difference measurements [4]. Fenton's
group switched from the vehicle-follower to the point-follower approach in 1971
because of the unavailability of radar sensors that would meet their performance
needs (f0.3 m accuracy over a range of 10- 100m) [4].
Caudill and Garrard explored this same information structure in considerable
574
STEVEN E. SHLADOVER
depth by means of analysis and simulation [119-1211. Unfortunately, because

they were studying the dynamics of a linear-induction-motor propelled vehicle,
they used a second-order vehicle model rather than a third-order model, thereby
losing all jerk information. In this work, the feedback of spacing and velocity
difference was augmented with further feedforward of the velocity of the
preceding vehicle. A PID controller was used for spacing regulation and PI
control for velocity regulation. They addressed the effects of acceleration or
thrust limiting in the control system, and showed how it could induce jumpresonance response unless the limit saturation was softened to a gain-change
nonlinearity. Garrard subsequently extended this work to consider the destabilizing effects of jerk limiting as well [122]. Olson and Garrard [I231 further
extended this to the development of two model-follower longitudinal controllers,
one of which used the spacing and velocity difference as inputs, while the other
added acceleration difference, which would put it in control structure 5.
Rouse and Hoberock used the same feedback variables for their. analysis of
vehicle following within platoons, but they retained a third-order vehicle
dynamic model, with a first-order lag for propulsion system dynamics
[I 24,1251. They adopted a "constant k-factor" vehicle follower spacing criterion
within their platoons, while assuming that the platoon as a whole would be
governed by a conventional block control system. The use of block control to
govern the speed of the platoon leader created significant difficulties, which
would seem to make it advisable to not try to apply this combination of control
systems.
In recent years, this feedback control structure has become extremely popular
for use in autonomous adaptive cruise control (AACC) systems (commonly
referred to as "Autonomous Intelligent Cruise Control" [AICC] based on the
terminology adopted within the PROMETHEUS program, but now considered
a less appropriate term because of the potential for the word "intelligent" to raise
expectations of performance to unrealistic levels). The interest within the
automotive and automotive supply industries in marketing automatic vehicle
following products and the recent advancements in ranging sensor technologies
have led to the explosion of activity here. Most of this activity has been focused
on the design of systems that would provide assistance to the driver to increase
comfort and convenience, but not necessarily automate all of the driving
functions.
Zhang [I261 defined several levels of longitudinal control being considered
within PROMETHEUS, including distance warning systems, AACC, and the
"light convoy", in which vehicles would be in communication with each other.
That paper defined concepts but did not include any test results. Benz and Zhang
[I271defined a modeling framework for evaluating the performance and safety of
AACC systems, but without showing any specific system design features. The
PROMETHEUS experience with these systems, which are considered the
primary success story of the entire program, is summarized in a two-part
paper by Becker et al. [128,129]. Significantly, the focus in that paper is on the
575
human interface issues rather than on vehicle dynamics or control law design.
Technical issues were addressed more specifically by Sala and Pressi [130], who
addressed the different spacing criteria that were considered for use in the
PROMETHEUS AACC work. This included the general categories that were
already mentioned here in Section 6.3, and provided more specifics on how the
safe following distances were to be estimated. This paper also defined strategies
for cooperative ACC, in which the vehicles are linked by communication systems
as well.
Work on AACC in Japan and the U.S. has lagged somewhat behind that in
Europe. Satoh and Taniguchi [131] described an AACC based on use of a laser
radar sensor and a controller designed to try to approximate normal driver
behavior. Their experiments on the test vehicle showed that drivers were quite
comfortable with the system's responses, but the paper gave no indication of
quantitative effects on traffic flow or the stability of this type of controller when
operated on multiple consecutive vehicles. Ioannou and his students have
applied this AACC control structure to the design of controllers for sequences
of fully automated vehicles operating in reserved lanes [132,133]. They have
shown that they can achieve asymptotic string stability with this simple feedback
structure by applying the constant (time) headway vehicle following criterion.
Their extensive controller design and simulation work has been followed with
some experiments implementing the controllers on test vehicles [134].
The largest body of technical papers on AACC has been produced by Paul
Fancher and his colleagues at the University of Michigan, working in cooperation with Leica, who developed the prototype hardware with which they have
been experimenting [135- 1381. They described their controller as fundamentally
a PI compensator with a limit on integrator wind-up [135]. Their infrared sensor
measures the range to the preceding vehicle, and they derive range rate from that
range information. They very carefully considered how to design the controller
for operation throughout the phase plane defined by range and range rate
measurements [136]. They evaluated the likely effect of the AACC on traffic flow
by means of simulations of deceleration maneuvers by strings of AACC vehicles.
The controller design focused on choosing the most appropriate vehiclefollowing time gaps (loosely speaking, headways) based on the control system
time constants, and concluded that the AACC would show asymptotic string
stability if the time gap is less than twice the control system time constant (lag)
for a constant-headway vehicle following mode of operation [137]. Experiments
with a variety of drivers in the equipped test vehicle highlighted the need to be
able to adjust performance.to reflect differing driver preferences [138].
6.3.3.Feedback of Spacing and Velocity DggPrence Relative to Preceding Vehicle

Plus External Velocity Comnrand (Structure 3 )
The asymptotic stability of a vehicle string can be improved by providing
additional common velocity command feedback to all of the vehicles. In this
576
STEVEN E. SHLADOVER
control structure, a roadside commanded velocity is added. Foster [139]

originally proposed the use of an absolute velocity control loop to provide
asymptotic stability to a constant-spacing vehicle following system. Other
control structures, to be addressed later, can provide stronger stabilization of
the constant-spacing systems.
Palmquist [140] proposed this structure in the context of an ACC in which the
traffic management center could command common speeds of all equipped
vehicles in order to harmonize traffic behavior and enhance safety by causing
vehicles to obey posted speed limits and traffic signals. The paper describes the
technical approach taken to implement this system and refers to experiments and
field trials that have already been conducted in Sweden, but does not present any
quantitative experimental results. The technology used here for roadside to
vehicle communication of speed commands could also be used for vehicle-tovehicle communication for cooperative cruise control.
6.3.4. Feedback of Spacing and Velocity Difference Relative to Preceding Vehicle
Plus Acceleration of Controlled Vehicle (Structure 4 )
Chiu, Stupp and Brown of the Applied Physics Laboratory investigated this
controller feedback structure in considerable depth in a series of papers in the late
1970s [141-1431. They defined a kinematic boundary to limit the approach of a
vehicle to its predecessor, based on possible anomalous motions of the predecessor. They also called attention to the limitations of linear time-invariant
controllers for short spacings if these controllers must also operate at larger
spacings. If the gains are chosen to be high enough for tight control at short
spacings, they tend to produce excessivecontrol action at larger spacings.This was
addressed by making the gains time-dependent during overtaking maneuvers and
including acceleration and jerk limits explicitly as saturation elements. They also
appear to have been the first to recognize the destabilizing effect of jerk limiting.
This structure has been applied more recently for vehicle-following controllers in Germany and Japan. Miiller and Nocker of Daimler Benz described an
AACC vehicle controller using an infrared range sensor and fuzzy control logic
to try to produce a driving experience that seems natural to the user within this
framework [144]. This was implemented on a test vehicle, and the paper included
the results of testing indicating that the fuzzy rules had not yet been fully
perfected. This structure was also applied by Fujita et al. of Honda [145] for a
collision-avoidance braking system. In this work, much attention was devoted to
defining the boundaries of accepta'ble operation in the spacing-closingrate phase
plane. The system was implemented on a test vehicle and evaluated in a series of
tests.
6.3.5. Feedback of the "Complete" State of the Preceding Vehicle and Controlled
Vehicle (Spacing and Velocity and Acceleration Differences) (Structure 5 )
This structure appears to have been used in only one paper, by Olson and
Garrard [123], where it was compared with another model-follower structure
577
without the acceleration feedback. The controller with this higher-order feedback also generates a higher-order control command (a jerk command as
compared to an acceleration command for the other controller). This controller
was shown to be vulnerable to instability because of the jerk-limiting saturation
nonlinearity.
Vehicle Plus Commanded Velocity of Preceding Vehicle (Structure 6 )
This structure was proposed by Tsugawa [146,147]as part of his concept of "soft
linked" vehicles. This concept was proposed for implementation using communications between vehicles rather than range or range rate sensing. Each vehicle
would need to know its own absolute position in order to avoid the need to
measure the spacing between them. This was originally proposed for very lowspeed use within factories or warehouses, with infrared communications between
successive vehicles, but then was extended to road vehicle applications. It was
simulated, but does not appear to have been implemented experimentally on
road vehicles.
Vehicle Plus Jerk of Preceding Vehicle (Structure 7 )
This structure was analyzed in depth by Pue of the Applied Physics Laboratory
in the late 1970s [148- 1511as a way of overcoming the awkwardness of the timevarying gains in the work by Chiu, Stupp and Brown. Pue developed a family of
rather complicated nonlinear controllers that explicitly incorporated the kinematic boundary of his colleagues and state constraints as well. His optimum and
one of his suboptimum controllers used this information structure, while he also
studied two other suboptimum controllers of structure 5. By means of simulations and describing function analysis he was able to demonstrate asymptotic
stability and good performance of this controller on a triple integration vehicle
model, with no propulsion system dynamics. Since this approach requires
communication of data that cannot be measured directly, these studies included
consideration of the needed data rates and the effects on performance of
communication imperfections (sampling, quantization and delays) [15 11.
6.3.8. Feedback of the Spacing and Velocity Dlfference Relative to Preceding

Vehicle Plus Velocity Dlfference Relative to Platoon Leader (Structure 8 )
This structure was proposed by Shladover [I521as a way of providing asymptotic
stability for constant-spacing vehicle following within a platoon of vehicles
operating very close together. The feedback from the platoon leader (which must
be provided by means of communication rather than direct sensing) serves as an
"absolute" reference for damping disturbances. This work demonstrated how to
overcome the destabilizing effects of both jerk and acceleration limiters in the
high-performance control loop needed to maintain tightly controlled spacings
between vehicles.
'
578
STEVEN E. SHLADOVER
6.3.9. Feedback of the "Complete" State of the Controlled Vehicle and its
Predecessor, Plus the Velocity andAcceleration of the Platoon Leader (Structure 9 )
This extension of feedback structure 8 was developed by researchers working in
the PATH Program, to provide more flexibility in defining the closed-loop
dynamics of the complete multi-vehicle coupled platoon with constant intraplatoon spacings. The addition of acceleration feedback makes it possible to
place the zeros as well as the poles of the closed-loop system transfer functions,
although it does require additional communication of data among the vehicles.
This has been applied to relatively simple third-order vehicle dynamic models
that incorporate some limited nonlinear effects [153,154] as well as models that
include explicit representations of powertrain dynamics [155-1571. The controller design approaches have included linear pole-placement [153,154] and
nonlinear sliding control [155- 1571.
These analysis and simulation studies demonstrated asymptotic stability of
the platoon of vehicles under relatively idealized conditions, as well as in the
presence of real-world imperfections such as process and measurement noise,
and quantization and sampling errors [153]. A companion study [I581 illustrated
the moderate loss of asymptotic stability that resulted from the loss of the
communication of lead vehicle information, again under idealized conditions,
while another [I591 showed the effects of communication delays, differences in
vehicle performance, and measurement noise.
This family of controller, building on the sliding control designs of Hedrick
and his colleagues described in 1155-1571, has been evaluated in extensive
experiments by PATH Program researchers, beginning with two vehicles and
extending to the control of four-vehicle platoons. Initial experiments were not
able to conclusively show asymptotic stability because of ranging sensor
limitations [160], but more recent experiments with an improved ranging sensor
have produced superior results, with excellent tracking accuracy and ride quality
[161,162]. The earlier experiments in this series illustrated the importance of
having fast throttle actuator response in order to achieve stability of vehicle
following (even with only two vehicles). The latest experiments have demonstrated the ability to operate four vehicles at a constant spacing of 4 m with
variations within 30cm, while executing a driving cycle including moderate
accelerations (0.1 g) and decelerations (0.05 g) on level roadway as well as on
grades of up to 3%. These experiments have been conducted using throttle
control but no brake control, because related studies of braking actuation have
shown the inadequate dynamic response of the brake actuator system that acts
on the brake pedal (thereby including brake booster dynamics) [163].
6.3.10. Feedback of the "Complete" State of the Controlled Vehicle and Its N
Predecessors (Structure 10)
This structure has been termed "Continuous Platooning" by Ren and Green
[164], whose analyses have indicated its advantages in responding to disturbances several vehicles ahead of the controlled vehicle (but not as far ahead as a
579
platoon leader). The communication burden depends on the number of vehicles

that would be included in the sub-string or sub-platoon of vehicles (N + I),
which is a design variable. This approach was analyzed much earlier by Athans
[165], who showed that use of sub-strings of only two or three vehicles could give
error correction performance almost as good as that achieved with full-state
feedback from the entire vehicle string or platoon (structure 11). Athans'
approach relied on optimal control design with full-state feedback, while
Garrard et. al. [166] applied full-state feedback within limited-length substrings
of vehicles and then used reduced-state feedback for controllir-g spacing between
the substrings. Both of these earlier studies were motivated by the desire to
reduce computational effort at a time when computing was still costly.
6.3.11. Full-State Feedback from all Vehicles (Structure I I )

This classic optimal control formulation was first proposed by Athans and his
associates for both continuous [I651 and sampled [167] cases. Optimal linear
regulator theory was used to choose feedback gains (for a design example based
on high-speed guideway transportation rather than road vehicles), and simulations showed good performance under normal operating conditions. Based on
the technological capabilities of the mid 1960s era of these studies, it was
assumed that measurements would be communicated from each vehicle to a
wayside control computer, saving on communication channel needs as well as
the number of (expensive) computers. Powner et a1 [168] addressed the same
problem using dynamic programming, claiming substantial computational
advantages for real-time implementations. Peppard and Gourishankar [I691
reduced the dimensions of the problem somewhat by eliminating the feedback
paths from vehicles behind the controlled vehicle.
6.3.12.Feedback of "Comp1ete"State Information of Controlled Vehicle, Platoon

Leader and Preceding and Trailing Vehicles (Structure 12)
This unconventional structure was proposed by Yang and Tongue [I701 as a way
of reducing the severity of the impacts that would occur between vehicles in a
close-formation platoon following a failure by an earlier vehicle. Known as
"back control", this is intended to cause the controlled vehicle to seek an
intermediate spacing between its predecessor and follower so that impacts
could be avoided to the extent possible, or else so that the impacts would be
comparable at the front and the rear. This was evaluated in simulation and
compared to the platoon controller of [154], showing improved ride quality and
platoon entry and exit operations. The authors have noted that by introducing
coupling of the controlled vehicle to both preceding and following vehicles, the
overall platoon dynamics are significantly more complicated than with the more
conventional controller and there are more opportunities for disturbances to
propagate.
580
STEVEN E. SHLADOVER
6.3.13. Other Controllers

There is an additional class of controllers that do not fit cleanly into the feedback
structures already described. These typically use limited measurement data, but
incorporate more complexity in a dynamic model of desired behavior that the
controlled vehicle is expected to follow. While a couple of model-follower
formulations that fit within the previously defined structures were described
by Olson and Garrard [123], others do not fit so neatly. For example, Sklar et a1
[171] defined a model-follower system based on building all of the complexity
into the model and then using a very simple but tight velocity regulator control
loop to force the vehicle to follow the model's trajectory. The model incorporated a kinematic constraint based on safe spacing considerations as well as other
features for jerk limiting and operations when there was no preceding vehicle
within range. Caudill and his students [I 72,1731 also studied model followers for
road vehicles, including mixes of light and heavy vehicles, with an emphasis on
the velocities of the controlled and preceding vehicles and the throttle setting of
the controlled vehicle. Youcef-Toumi et a1 [174] described yet a different kind of
model following system that relies only on a direct measurement of range to
implement an adaptive cruise control. This was tested in the laboratory on a 115
scale model vehicle.
None of these model-follower systems have been implemented on full-scale
test vehicles, but simulations of their performance under idealized conditions
have predicted good performance.
Vehicle-follower control design is always based on the assumption that there
is a preceding vehicle to be followed. However, in cases where no other vehicle is
"visible" in front of a controlled vehicle, it needs to have suitable control laws to
follow. This issue was initially addressed by Rouse and Hoberock [124,125] by
assuming use of block control for the leader of a platoon, as already mentioned
in Section 6.3.2. More recently, Godbole and Lygeros [175] have proposed the
design of a platoon leader control law that includes multiple modes of operation,
based on more sophisticated consideration of the state of the platoon leader
relative to the preceding platoon.
6.4. Sensors for Longitudinal Control
The literature on sensors is vast and generally distinct from the vehicle dynamics
and control literature. There have been a few papers directly focused on ranging
sensors for longitudinal control, especially for radar, which is the most
commonly recommended sensor modality. Grimes [I761 gave a description of
his automotive radar system which is surprisingly detailed considering the
competitive pressures that have led other technology developers to be quite
secretive about their technology. The European DRIVE program's project
SMILEK (Short range Microwave Links for European Roads) sponsored the
one published comparison of the available forward-looking radar sensor systems
[177]. Although this incorporates substantial information, the developments in
automotive radar have been progressing at such a rapid pace that this review is
58 1
already somewhat dated (presented at the February 1991 DRIVE Conference).

A more recent discussion of Philips' radars for adaptive cruise control was
presented by Stove [178], but this paper does not contain much detailed technical
information. Kikuchi et. al. [179] provided somewhat more information about
Honda's development of a laser radar sensor system.
6.5. Summary of State of Current Knowledge on Longitudinal Control
The wealth of research that has been done on longitudinal control systems has
meant that many diverse approaches have been investigated analytically and in
simulation. However, very few of these have been validated experimentally, in
the presence of real-world sensor and actuator imperfections, parameter variations, noise and external force loadings. This means that attention needs to be
focused on the limited subset of the research that has involved such validation, in
order to avoid being misled by unsupportable claims.
The key issues that must be kept in mind are:
a In light of modern developments in computers, sensors, communication
and control technology, interest has focused much more heavily on vehiclefollower longitudinal control systems than on point-follower or fixed block
systems in recent years. Although these systems require more complicated
technology to be implemented on the vehicles than the others, their
advantages in terms of overall system capacity (vehicles per hour), fault
tolerance, and total system cost outweigh that concern.
a The longitudinal dynamics of light-duty vehicles powered by spark ignition
engines with electronic fuel injection do not appear to be limitations to the
achievement of high performance longitudinal control systems, with
smooth and precise control of spacings between vehicles.
a A wide variety of possible feedback control structures have been proposed
and evaluated over the years, although only a few of them have been
verified experimentally. A few of these structures can be implemented on
autonomous vehicles (using vehicle-mounted sensors only, without communication), but most of them require communication of state information
from other vehicles.
a There are limits to the performance that can be achieved at each level of
feedback information (or control structure), and higher levels of performance generally require more information.
a Vehicle-follower systems should be designed to have asymptotic string
stability for an arbitrary length string of vehicles. Designs that are stable for
one vehicle following another are not necessarily asymptotically stable for
larger numbers of consecutive vehicles. Asymptotic stability needs to be
assured under all realistic abnormal operating conditions, including
communication delays, sensor noise, sampling and quantization errors,
parameter variations, etc. ,
a Different criteria can be applied to determine the relationship between
582
STEVEN E. SHLADOVER
longitudinal spacing and vehicle speed. These are generally classified as

constant spacing, constant headway (time) and constant safety factor. One
of the most important general findings for longitudinal control system
performance, as reported by Swaroop et. al. [180], is that asymptotic
stability can be provided for constant-headway vehicle following using
autonomous sensing (without communication), while constant-spacing
systems require communication of additional information to achieve
asymptotic stability.
Dynamic response of actuation systems (engine and brake control systems)
can be a serious limitation to longitudinal control system performance.
Fast response of these subsystems is needed to ensure that high-precision
constant-spacing vehicle follower control systems perform effectively and
with asymptotic stability.
The selection of the desirable spacing for vehicles remains challenging and
controversial, based on differing perceptions of the relative safety of
different spacings under various operating conditions. This is almost as
much a philosophical issue as it is a technical issue.
The largest technical issue in the design of regulation-layer controllers for
longitudinal control is ensuring robustness with respect to parameter
variations, external force loadings and sensor anomalies under the full
range of real-world operating conditions. This will require extensive
experimental work on full-scale vehicles in realistic operating environments.
Ranging sensor refinement is probably the largest remaining challenge in
longitudinal control technology, particularly in ensuring that the sensor
always finds the right target vehicle or obstacle.
7. REVIEW O F INTEGRATED LATERAL/LONGITUDINAL CONTROL

Although the bulk of AVCS research to date has concentrated on either lateral or
longitudinal control, there have been some projects that have attempted to
integrate both in a combined vehicle control system. These have ranged from
warnings to partial control, intermittent control (for emergency conditions) and
then full-time control.
The warning and control assistance systems have been reported within the
past year by researchers from Japanese and German automotive companies.
Sekine et al [I811 reported on Honda's development and testing of a system
designed to match driving speed to the local curve radius, based on curvature
information derived from the map database of the vehicle's navigation system.
This includes a warning to the driver if the lateral acceleration exceeds 0.2 g and
an automatic deceleration if it exceeds 0.4g. Butsuen et a1 [I821 described
Mazda's collision avoidance system, which applies the brakes automatically
based on multiple sensor inputs. These include a scanning laser radar to detect
583
vehicles or other obstacles, a video image processing system to detect lane

delineations and a road surface condition sensor. Even though the control is only
applied longitudinally (braking) the sensing function includes both lateral (lane
position) and longitudinal (ranging to obstacle).
Both Porsche and BMW have chosen to emphasize driver assistance rather
than complete vehicle control in their research within the PROMETHEUS
Program. Reichart has described BMW's work on combined lateral and longitudinal driver control assistance, which is implemented by providing torque and
force feedback to the driver through the steering wheel and accelerator pedal
respectively [183,184]. This approach requires effective modeling of the driver's
responses to these stimuli, introducing a significant human factors complication
into the dynamics of the control system. The systems have been tested by drivers
of equipped vehicles in real traffic, and the results have been reported to be
encouraging. Andrews et al [185] have described Porsche's work on optimizing
the use of the tire friction ellipse for controlling the vehicle under extreme
conditions. This has included an innovative acoustic method of estimating tire/
road friction, and integration of that friction estimate with driver warning and
cruise control systems. This means that the vehicle can give the driver a warning
that he is on the verge of exceeding the friction potential of the tires, or it can give
the driver force feedback through the accelerator pedal ("interactive cruise
control"), or it can control the vehicle's speed directly (adaptive cruise control).
Research on integrated lateral/longitudinal control for fully automated
vehicles has been conducted by both academic and industrial researchers.
From the academic world, Kehtarnavaz [186] described the use of a binocular
image processing system to enable a vehicle to follow a leading vehicle equipped
with a bright target light. This vehicle following included both steering and speed
control. Sheikholeslam and Desoer [187] described a modeling and simulation
study of combined lateral/longitudinal control using relatively simple vehicle
models, and without significant coupling of the lateral and longitudinal motions.
Pham [I881 extended this to more sophisticated models and control laws,
showing the merits of coupling the motions explicitly in simulations. Pomerleau
et al. [189] reviewed the research that he and his colleagues at the Robotics
Institute of Carnegie-Mellon University have performed on autonomous
unmanned ground vehicles, including lane following, vehicle following and
parking, using machine vision for sensing. This research has emphasized the
artificial intelligence issues associated with image processing much more than the
vehicle dynamics or control issues.
The industrial research on integrated lateral/longitudinal control for fully
automated vehicles appears to have progressed further, with some significant
experimental implementations. Mazda's development of fully-automated test
vehicles using image processing for lane following and obstacle detection has
been described in papers beginning in 1990 [I901 and continuing more recently
with the inclusion of results from testing with drivers sitting in the test vehicles
while they drive themselves on test tracks [191]. Similar work by Mitsubishi was
584
STEVEN E. SHLADOVER
also reported as early as 1990 [192], with an emphasis on design to optimize ride
comfort for the vehicle passengers in automatic lane change maneuvers.
Toyota's research, on fully automated driving concentrated initially on testing
of unmanned vehicles on the company's test track, rather than on public roads
[193]. More recently, they have extended their work to the development of
systems that could be applied to automate the driving of customers' vehicles on
suitably equipped roads (using inductive cable guidance combined with stereo
vision for lane detection and a laser radar for obstacle detection/vehcle
following) [194]. Probably the most sophisticated such development to date
has been the work of Daimler Benz on the VITA I1 test vehcle, an automobile
equipped with 12 video cameras and special-purpose image processing computers to handle all aspects of regulation-layer driving [195]. This vehicle incorporates lane tracking, vehicle following and forward and side obstacle detection
and avoidance, and has been driven under automatic control on public highways
in Germany.
8. REVIEW O F HIGHER-LEVEL AVCS CONTROL ISSUES

Attention has been focused thus far on the regulation-layer functions of
automatic lateral and longitudinal control. However, it is important for fully
automated systems in particular to perform higher-level functions as well,
recognizing the trajectories that they need to follow to negotiate the surrounding traffic safely, as well as having means for optimizing traffic flows of
automated vehicles.
The coordination layer issues were first addressed by Hsu et. al. [196], who
devised a set of protocols to govern the fundamental maneuvers that vehicles can
perform in cooperation with their neighbors (lane change, join a platoon or split
from a platoon). The logic of these maneuvers (including the inter-vehicle
communications) was defined in terms of finite state machines, and the correctness of the maneuvers was then proven [197]. Some of the challenges of
integrating this discrete-event representation with the continuous regulation
layer were identified by Lygeros and Godbole [198], as part of a continuing
sequence of work that remains far from complete. Rao and Varaiya used a micro
simulation to evaluate the implications of these coordination layer maneuvers on
lane capacity of an automated system both with and without communication
among the vehicles [199,200]. Niehaus and Stengel adopted an expert system
approach to defining the maneuvers that an autonomous vehicle would make,
based on the actions of other vehicles in its vicinity that it can perceive by means
of machine vision [201-2031. In contrast to the previous papers, this study
assumed no communication is available among the vehicles. An analogous
approach to implementing defensive driving rules for a merging maneuver was
proposed by Pilutti et a1 [204], with the intended application to a driver warning
system rather than a fully automated control system.
585
Above the coordination layer is the link layer, where the relevant variables are
associated with the flows of vehicles rather than the motions of individual
vehicles. The earliest treatment of the issues that arise at this layer appears to be
by Karaaslan et a1 [205], who addressed the capacity of an automated lane and
the ways in which wayside recommendations of speed could be used to attenuate
flow disturbances. These issues were explored in much greater depth by Rao and
Varaiya, who developed heuristics for assigning vehicles to lanes within an
automated roadway facility under both normal and incident conditions
[206,207]. These papers showed very good system performance with the
heuristic control approach. Ramaswamy et a1 [208] have taken .a different
approach to automatic lane assignment, based on a centralized optimization
of trajectories that appears to be very communication intensive. Peng has
adopted a more traditional optimal control approach to the link-layer traffic
management problem [209].
The highest layer of the fully automated control system is the network layer.
Some of the issues at this layer were addressed in studies of automated guideway
transit systems in the 1970s, and Fenton's group at OSU addressed them for
automated highways at that time as well [4]. Those earlier studies must now be
considered obsolete because of the dramatic changes in computer hardware and
software and communications technologies that have occurred since then. The
first tentative modern approach to the issues encountered at this layer appears to
have been by Tsao [210], but much more remains to be done here.
9. CONCLUDING REMARKS
AVCS is an example of a field of activity in which we need to repeatedly remind
ourselves that there are few completely new ideas. There is a heritage of more than
thirty years of applicable research upon which to build. During this time, much
has been learned about concepts and technical approaches, both favorable and
unfavorable. That body of knowledge has been documented to a considerable
extent, and is available to researchers who are interested in working on AVCS.
This paper has attempted to summarize the highlights of what has been
accomplished in AVCS. It is not meant to be an exhaustive survey of AVCS, and
indeed it does not even address many aspects of AVCS. However, it should
provide reasonably complete coverage of the state of knowledge about the
vehicle dynamics and control aspects of AVCS. Readers who take the time to
consult the references cited herein should find that a worthwhile investment, one
that will save them much time and wasted effort.
ACKNOWLEDGMENT
This work was performed as part of the California PATH Program of the
586
STEVEN E. SHLADOVER
University of California, in cooperation with the State of California Business,

Transportation, and Housing Agency, Department of Transportation; and the
United States Department of Transportation, Federal Highway Administration.
The contents of this report reflect the views of the author who is responsible
for the facts and the accuracy of the data presented herein. The contents do not
necessarily reflect the official views or policies of the State of California. This
report does not constitute a standard, specification, or regulation.
REFERENCES
I . Gardels, K.: Automatic car controls for electronic highways. General Motors Research
Laboratories Report GMR-276, 1960.
2. Zworykin, V.K., and Flory, L.E.: Electronic control of motor vehicles on the highways. Highway Research Board Proceedings, 37th Annual Meeting, 1958.
3. Barrick, D.: Automatic Steering Techniques. IRE Inter. Conv. Rec. 10, Part 2 (1962), pp. 166178.
4. Fenton, R.E., and Mayhan, R.J.: Automated highway studies at the Ohio State University An overview. IEEE Trans. on Veh. Tech. 40,l (1991), pp. 100-1 13.
5. Hanson, M.E. (Ed.): Project METRAN: An Integrated, Evolutionary Transportation System
for Urban Areas, MIT Press, 1966.
6. Tomorrow's transportation: new systems for the urban future. U.S. Dept. of Housing and
Urban Development, 1968.
7. Dual mode transportation. Proc., conference conducted by Transportation Research Board,
May 29-31, 1974 TRB Special Report 170, 1976.
8. TRW Systems Group: Automated Highway Systems, December 1969, 572 pp. (PB 191 696).
9. Calspan Corp.: Practicability of automated highway systems. Report DOT-FH-11-8903,
1977.
.. . ..
10. Bender, J.G., et al.: System studies of automated highway systems. General Motors Transportation Systems Center Report EP-81041A. 1981; FHWA Report No. FHWAIRD-821
003, 198i.
11. Bender. J.G.: An overview of svstems studies of automated hiehwav svstems. IEEE Trans.
~ e hTech.
.
40,l (1991), p i . 82-99.
12. Smisek, R.R., and Harder, G.A.: Automated control of guideway transit vehicles. 29th IEEE
Vehicular Technology Conference, Arlington Heights, IL, March 1979, pp. 57-64.
13. Anderson, J.E., et al. (Eds.): Personal rapid transit. Institute ofTechnology, University of Minnesota, 1972.
14. Anderson, J.E. (Ed.): Personal rapid transit 11. University of Minnesota, 1974.
15. Gary, D.A. (Ed.): Personal rapid transit 111. University of Minnesota, 1976.
16. Advanced Transit Association: International Conference on Advanced Transportation and
Urban Revitalization, Indianapolis, IN, May 1978.
17. Cardew, K.H.F.: The Automatic Steering of Vehicles - An Experimental System Fitted to a
DS Citroen Car, Road Research Laboratory Report RRL LR 340, Crowthorne, England, 1970, 29 pp. (PB 196 983).
18. Oshima, R., et al.: Control system for automobile driving. Proc. Tokyo IFAC Symposium
(1965), pp. 347-357.
19. Fujioka, T., Yoshimoto, K., and Takaba, S.: A case study on an automated driving highway
system in Japan. Transportation Research Board 72nd Annual Meeting, Washington,
D.C., January 1993.
20. Hesse, R.: Cabinentaxi - A personal public transportation system. Transp. 1,3 (1972), pp. 321329.
21. Ishii, T., et al.: The control system of CVS using the two-target tracking scheme. personal
Rapid Transit 11, Dept. of Audio Visual Extension, University of Minnesota (1974),
pp. 325-334.
- - -
STATE O F DEVELOPMENT O F AVCS
587
22. Maury, J.P.: The ARMIS PRT system. SAE Paper 740143, 1974.
23. Broqua, F., Lerner, G., Mauro, V., and Morello, E.: Cooperative driving: basic concepts and a
first assessment of "intelligent cruise control" strategies. Proc. Advanced Telematics in
Road Transport, Vol. 11, DRIVE Conference, Brussels, February 1991, pp. 908-929.
24. de Saint Blancard, M., Dubuisson, B., and Rombaut, M.: PRO-LAB2: A vehicle with driving
assistance. Proc. First World Congress on Applications of Transport Telematics and
Intelligent Vehicle-Highway Systems, Paris, November-December 1994. pp. 1871 - 1878.
25. Saxton, L.: Mobility 2000 and the roots of IVHS. IVHS Rev. (Spring 1993), pp. 11-26.
26. Shladover, S.E., Bushey, R., and Parsons, R.E.: California and the roots of IVHS. IVHS Rev.
(Spring 1993), pp. 27-34.
28. Tsugawa, S., Watanabe, N., and Fujii, H.: Super smart vehicle system - its concept and preliminary works. Proc. Vehicle Navigation and Information Systems Symposium, Dearborn, MI, October 1991, pp. 269-277 (SAE Publication P-253).
29. Nakamura, M., Kawashima, H., Yoshikai, T., and Aoki, K.: Road-vehiclecooperation driving
system. Proc. 1994 Vehicle Navigation and Information Systems, Yokohama. Japan,
August-September 1994, pp. 425-430.
30. Leasure, W.A., and Burgett, A.L.: NHTSA's IVHS collision avoidance research program: strategic plan and status update. Proc. First World Congress on Applications of Transport
Telematics and Intelligent
Paris, November-December 1994,
- Vehicle-Highway
- Systems,
pp. 22 16-2223.
31. Shladover. S.E.: The California PATH Droaram of IVHS research and its aporoach to vehiclehighhay automation. Proc. lnteliigent Vehicles 92 Symposium, ~ e t i d i t MI,
, June-July
1992, pp. 347-352.
32. Chang, K.S., Hedrick, J.K., Zhang, W.-B., Varaiya, P., Tomizuka, M., and Shladover, S.E.:
Automated highway system experiments in the PATH program. IVHS J. 1,1 (1993),
pp. 63-87.
33. Saxton, L., and Bishop, J.R.: The United States DOT automated highway system (AHS) program. Proc. First World Congress on Applications of Transport Telematics and Intelligent Vehicle-Highway Systems, Paris, November-December 1994, pp. 1952-1958.
34. Bishop, J.R., and Stevens, W.: Results of precursor systems analyses of automated highway
systems. Proc. First World Congress on Applications of Transport Telematics and Intelligent Vehicle-Highway Systems, Paris, November-December 1994, pp. 1959-1966.
35. National Program Plan for Intelligent Transportation Systems, Final Draft, ITS America,
November 1994.
36. Varaiya, P., and Shladover, S.E.: Sketch of an IVHS Systems Architecture, Proc. Vehicle Navigation and Information Systems Conference, Paper No. 912838, Dearborn, MI, October
1991, pp. 909-928.
37. Zhang, W.-B., Shladover, S.E., Hall, R.W., and Plocher, T.: Functional definition of automated highway systems, TRB Paper 940988, 73rd Annual Meeting, January 1994,
Washington, D.C.
38. Transportation Research Institute, Carnegie-Mellon University, Urban Rapid Transit Concepts and Evaluation, TRI Research Report #I, Pittsburgh, 1968, 241 pp.
39. Bidwell, J.B., and Cataldo, R.S.: Electronic chauffeur may guide car of future. SAE J., (September 1958), pp. 64-67.
40. Electronic control of motor vehicles on the highways. Proc. Highway Research Board 37th
Annual Meeting, 1958, pp. 436-451.
41. Flory, L.E., Gray, Morey, Pike, and Caulton: Electronic techniques in a system of highway
vehicle control. RCA Rev. 23,3 (1962), pp. 293-310.
42. Penoyre, S.: Dual mode road vehicles. Traff. Eng. Contr. 13,s (1971), London, pp. 189-191.
43. Road Research 1971. Annual Report of the Road Research Laboratory, London: Her Majesty's Stationery Office, 1972, 120 pp.
44. Fenton, R.E., Olson, K.W., and Bender, J.G.: Advances Toward the Automatic Highway.
Highway Research Record #344, 1971, pp. 1-20 (including commentary by Wilkie,
Lenard, and Segal).
45. Olson, K.W., Sapp, and Flaig: A System for Automatic Vehicle Lateral Guidance, Highway
Research Record #275, 1969, pp. 1-1 1.
46. Olson, K.W.: Wire-reference configurations in vehicle lateral control. IEEE Trans Veh. Tech.
VT-26,2 (1977), pp. 161-172.
588
STEVEN E. SHLADOVER
47. Fenton, R.E., Melocik, G.C., and Olson, K.W.: On the steering of automated vehicles: Theory
and experiment. IEEE Trans. Auto. Control AC-21,3 (1976), pp. 306-3 15.
48. Fenton, R.E., and Selim, 1.: On the optimal design of an automotive lateral controller. IEEE
Trans. Veh. Tech. VT-37,2 (1988), pp. 108-1 13.
49. Tsugawa, S., Yatabe, T., Hirose, T., and Matsumoto, S.: An automobile with artificial intelligence. Proc. Sixth International Joint Conference on Artificial Intelligence, Vol. 2,
Tokyo, Japan, August 1979, pp. 893-895.
50. Tsugawa, S.: Vision-based vehicles in Japan: the machine vision systems and driving control
systems. Proc. IEEE International Symposium on Industrial Electronics, Budapest,
Hungary, June 1993.
51. Spangler, R.A., and Snell, F.M.: A New Vehicle Guidance and Speed Control System, Highway Research Bulletin #338, September 1962, pp. 20-27.
52. Driverless Bus Tested. Automotive Industries, 125,9 (1960, p. 29.
53. Richards, 6.: New Movement in Cities, London: Studio Vista, 1966, 96 pp.
54. O'Rourke, C.O.: Electronic trail-finding. Contr. Eng. 4,5 (1957), pp. 117-1 19.
55. Pasternack, S.: Modern Control Aspects of Automatically Steered Vehicles, DOT-TSC-OST72-3, Technical Note, December 1971, 16 pp.
56. Pasternack, S.: Some modern control aspects of automatically steered rubber tired vehicles,
ASME Paper No. 73-ICT-54, Second lntersociety Conference on Transportation, Denver, September 1973.
57. Bonderson, L.S.: Optimal lateral control for dual-mode vehicles. Dual-Mode Conference,
Washington, D.C., May 1974.
58. Shladover, S.E.: Performance limits of automatic steering controllers. Gary, D.A. (Ed.): Personal Rapid Transit 111. University of Minnesota, Minneapolis, M N , 1976, pp. 233-245.
59. Shladover, S.E., Wormley, D.N., Richardson, H.H., and Fish, R.: Steeringcontroller design for
automated guideway transit vehicles, J. Dyn. Sys. Meas. Contr. 100,l (1978), pp. 1-8.
60. Shladover, S.E., Fish, R., Wormley, D.N., and Richardson, H.H.: Performance of steering
controllers for automated guideway transit vehicles. Proc. 5th Vehicle System Dynamics
- 2nd International Union of Theoretical and Applied Mechanics Symposium:
Dynamics of Vehicles on Roads and Tracks, Vienna, September 1977. Swets and Zeitlinger B.V., Amsterdam, pp. 127-135.
61. Christ, H., Darenberg, W., Panik, F., and Weidemann, W.: Automatic track control of
vehicles. Theory and experiment. Slibar, A,, and Springer, H. (Ed.): The Dynamics of
Vehicles on Roads and Tracks. Swets & Zeitlinger, Amsterdam, 1978, pp. 145-164.
62. Darenberg, W.: Automatische spurfiihrung von Kraftfahrzeugen. Automobil-Industrie (February 1987).
63. Weidemann, W.: Electronic lateral control of vehicles. SAE Technical Paper Series, No.
881 166, SAE, Warrendale, PA, Future Transportation Technology Conference and
Exposition, San Francisco, CA, August 1988.
Computer vision applied to vehicle operation. SAE Technical Paper Series,
64. Metzler, H.-G.:
No. 881 167, SAE, Warrendale, PA, Future Transportation Technology Conference
and Exposition, San Francisco, CA, August 1988.
65. Dickmanns, E.D., and Zapp, A.: A curvature-based scheme for improving road vehicle guidance by computer vision. Mobile Robots. Proc. SPIE, Vol. 727 (1986), pp. 161-168.
66. Dickmanns, E.D., and Zapp, A,: Autonomous high-speed road vehicle guidance by computer
vision. Tenth IFAC World Congress, Munich, 4 (1987), pp. 232-237.
67. Graefe, V.: Vision for intelligent road vehicles. Proc. Intelligent Vehicles '93 IEEE Symposium,
Tokyo, July 1993, pp. 135-140.
68. Masaki, I. (Ed.): Vision-Based Vehicle Guidance, Springer Verlag, New York, 1992.
, ~ u n e1992.
69. Proc. intelligent Vehicles '92 Symposium, IEEE, ~ e t r o sMI,
70. Proc. lntellieent Vehicles '93 Svmoosium. IEEE. Tokvo. Julv 1993.
71. Peng, H., an; Tomizuka, M. vehicle lateral control for-highway automation. Proc. 1990 American Control Conference, San Diego, May 1990, pp. 788-794.
72. Peng, H., and Tomizuka, M.: Preview control for vehicle lateral guidance in highway automation. Proc. 1991 American Control Conference, Boston, MA, June 1991, pp. 3090-3095.
73. Peng, H., Zhang, W.-B., Shladover, S.E., Tomizuka, M., and Arai, A,: Magnetic-marker-based
lane keeping: a robustness experimental study. SAE Congress, Detroit, MI, February
1993, SAE Paper No. 930556, pp. 127-132.
STATE O F DEVELOPMENT OF AVCS
589
74. Peng, H., Zhang, W.-B., Tomizuka, M., and Shladover, S.E.: A reusability study of vehicle
lateral control system. Veh. Sys. Dyn. 23,4 (1994), pp. 259-278.
75. Hessburg, T., and Tomizuka, M.: Fuzzy logic control for lateral vehicle guidance. IEEE Contr.
Sys. Mag. 14,4 (1994), pp. 55-63.
76. Chee, W., and Tomizuka, M.: Lane change maneuver for AHS applications. Proc. International Symposium on Advanced Vehicle Control 1994, Japan, October, 1994, p. 420425.
77. Narendran, V.K., and Hedrick, J.K.: Autonomous lateral control of vehicles in an automated
highway system. Veh. Sys. Dyn. 23,4 (1994), pp. 307-324.
78. Fujioka, T., and Suzuka, K.: Lateral autonomous driving by sliding control. Proc. International Symposium on Advanced Vehicle Control 1994, Japan, October, 1994, pp. 432437.
79. Ackermann, J., and Sienel, W.: Robust control for automatic steering. Proc. 1990 American
Control Conference, San Diego, May 1990, pp. 795-800.
80. Guldner, J., Utkin, V.I., and Ackermann, J.: A sliding mode control approach to automatic car
steering. Proc. American Control Conference, Baltimore, MD, June 1994, pp. 19691973.
81. Ackermann, J., Guldner, J., Sienel, W., Steinhauser, R., and Utkin, V.I.: Linear and nonlinear
controller design for robust automatic steering. IEEE Trans. on Contr. Sys. Tech. (1994).
82. Patwardhan, S., Tomizuka, M., Zhang, W.-B., and Devlin, P.: Theory and experiments of tire
blow-out effects and hazard reduction control for automated vehicle lateral control
system. Proc. International Symposium on Advanced Vehicle Control 1994, Japan,
October, 1994, pp. 426-43 1.
83. Smith, D.E., and Starkey, J.M.: Effects of model complexity on the performance of automated
vehicle steering controllers: controller development and evaluation. Veh. Sys. Dyn. 23,8
(1994), pp. 627-645.
84. Kushiro, I., Kawakami, S., and Saitou, R.: Crosswind feedforward control by active front
steering. Proc. International Symposium on Advanced Vehicle Control, Yokohama,
Japan, September 1992, pp. 402-406.
85. Adachi, T., Butsuen, T., and Kamiyama, S.: Application of real-time simulation to the lanekeeping system. Proc. 1994 Vehicle Navigation and Information Systems, Yokohama,
Japan, August-September, 1994, pp. 415-418.
86. Lemarchand, C., and Coffin, F.: Conception method of man machine interface for an automotive vehicle transversal driving assistance system. Proc. First World Congress on Applications of Transport Telematics and Intelligent Vehicle-Highway Systems, Paris,
November-December 1994, pp. 1820-1827.
87. Mayhan, R.J., and Bishell, R.A.: A two-frequency radar for vehicle automatic lateral control.
IEEE Trans. Veh. Tech. VT-3 1,l (1982), pp. 32-39.
88. Giubbolini, L., and Cugiani, C.: Radar systems for the control of the lateral position of the
vehicle traveling along a motonvay. Proc. International Symposium on Automotive
Technology and Automation, Aachen, Germany, 1993, pp. 689-696.
89. Cugiani, C., and Giubbolini, L.: Millimetre wave radar sensor for the highway global positioning of a vehicle. Proc. 1994 Vehicle Navigation and Information Systems, Yokohama,
Japan, August-September, 1994, pp. 409-414.
90. Johnson, B., et al.: A lateral guidance radar module. Proc. International Symposium on Automotive Technology and Automation, Aachen, Germany, 1993, pp. 697-703.
91. Bishel, R.A.: Dual-mode truck: automated and manual operation. SAE Future Transportation
Technology Conference, 1994.
92. Johnston, A.R., Assefi, T., and Lai, J.Y.: Automated vehicle guidance using discrete reference
markers. IEEE Trans. Veh. Tech. VT-28,l (1979), pp. 95-106.
93. Yasui, Y., and Margolis, D.L.: Lateral control of automobiles using a look-ahead sensor. Proc.
International Symposium on Advanced Vehicle Control, Yokohama, Japan, September
1992, pp. 292-297.
94. Zhang, W.-B., and Parsons, R.E.: An intelligent roadway reference system for vehicle lateral
guidance/control. Proc. 1990 American Control Conference, San Diego, May 1990,
pp. 28 1-286.
95. Chu, K.-C.: Decentralized control of high-speed vehicular strings. Trans. Sci. 8 (1974), pp.
361-384.
590
STEVEN E. SHLADOVER
96. Rahimi, A., Hajdu, L.P., and Macomber. H.L.: Safety analysis for automated transportation
systems. Trans. Sci. 7,l (1973). pp. 1-17.
97. Stepner, D.E., Hajdu, L.P., and Rahimi, A.: Safety considerations for personal rapid transit.
Personal Rapid Transit, Institute of Technology, University of Minnesota, 1972, pp.
457-472.
98. Pitts, G.L.: Augmented Block Guidance for Short-Haul Transportation Systems. Johns Hopkins University Applied Physics Laboratory Report APL JHU CP 019/TPR 023, September 1972, 178 pp. (PB 214 391).
99. Pitts, G.L., and Hinman, E. Augmented block guidance - a control concept for automated
urban transit, APL Tech. Dig., 12,1 (1973), pp. 2-10.
100. Pitts, G.L.: A collision avoidance control law for fixed block, short-headway transportation
systems. AIAA Paper 71-942, Guidance, Control and Flight Mechanics Conference,
Hempstead, NY, August 1971.
101. Whitney, D.E., and Tomizuka, M.: Normal and emergency control of a string of vehicles by
fixed reference sampled - data control. Personal Rapid Transit, Institute of Technology, University of Minnesota, 1972, pp. 383-404.
102. Smith, P.G.: Discrete-time longitudinal control of dual mode automobiles. ASME Paper 73ICT-77. 12 pp.
103.' Hinman, E . J . . ( E ~ :Command
~
and control studies for personal rapid transit, program status,
1974. Johns Hopkins University, Applied Physics Laboratory Report APLIJHU CP 0391
TPR 030, April 1975, 178 pp.
104. Garrard, W.L.: Longitudinal control for automated guideway transit vehicles. University of
Minnesota, Report UMTA-MN-I 1-0002-77-2, July 1977, 107 pp.
105. Fenton, R.E., and Chu, P.M.: On vehicle automatic longitudinal control. Trans. Sci. 11,l
(1977), pp. 73-91.
106. Wilkie, D.: Moving cell control scheme for automated transportation systems. Trans. Sci. 4
(1970), pp. 347-364.
107. Fling, R.B., and Olson, C.L.: An integrated concept for propulsion, braking, control and
switching of vehicles operating at close headways. Personal Rapid Transit. Institute of
Technology, University of Minnesota, 1972, pp. 361-382.
108. Garrard, W.L., and Kornhauser, A.: Design of optimal feedback systems for longitudinal control of automated transit vehicles. Trans. Res. 7.2 (1973), pp. 125-144.
109. Garrard, W.L., and Kornhauser. A,: Use of state observers in the optimal feedback control of
automated transit vehicles. J. Dyn. Sys. Meas. Contr. 95,2 (1973), pp. 220-227.
1 10. Kornhauser, A,, et al.: Optimal sample-data control of PRT vehicles. Personal Rapid Transit
11. Department of Audio Visual Extension, University of Minnesota, 1974, pp. 359-366.
1I I. Yang, S.C., and Garrard, W.L.: A low sensitivity modem approach to the longitudinal control
of automated transit vehicles. J. Dyn. Sys. Meas. Contr. 96,2 (1974), pp. 218-228.
112. Evans, R.R., and Whitten, R.P.: Analysis and demonstration of position error headway protection for PRT systems. Personal Rapid Transit [I. Department of Audio Visual Extension, University of Minnesota, 1974, pp. 335-348.
113. Whitten, R.P., and Wright, R.D.: Analysis of short ramps for dual-mode and PRT stations.
Paper presented at First International Conference on Dual-Mode Transportation,
Washington, D.C., May 1974, Transportation Research Board.
1 14. Whitten, R.P.: Short station ramps for AGT systems. IEEE Trans. Veh. Tech. VT-28,l (1979),
pp. 63-70.
115. Brown, S.J.: Design considerations for vehicle state control by the point-follower method. Personal Rapid Transit 11. Department of Audio Visual Extension, University of Minnesota, 1974, pp. 381-390.
116. Bender, J., and Fenton, R.: On the flow capacity of automated highways. Trans. Sci. 4 (1970),
pp. 52-63.
117. Fenton, R.E., et al.: One approach to highway automation. IEEE Proc. 56,4 (1968), pp. 556566.
1 18. Bender, J., Fenton, R., and Olson, K.: An experimental study of vehicle automatic longitudinal
control. IEEE Trans. Veh. Tech. VT-20,4 (1971), pp. 1 14-123.
119. Garrard, W.L.: Longitudinal control for automated guideway transit vehicles. University of
Minnesota, Report UMTA-MN-I 1-0002-77-2, July 1977, 107 pp.
120. Caudill, R.J., and Garrard, W.L.: Vehicle follower longitudinal control for automated guide-
59 1
way transit vehicles. University of Minnesota, Report UMTA-MN-11-0002-77-1,

February 1977, 265 pp.
121. Caudill, R.J., and Garrard, W.L.: Vehicle follower longitudinal control for automated transit
vehicles. J. Dyn. Sys. Meas. Contr. 99,4 (1977), pp. 241-248.
122. Garrard, W.L.: Effects of jerk limiting on the stability of automated transit vehicles. J. Dyn.
Sys. Meas. Contr. 100,4 (1978), pp. 298-301.
123. Olson, D.E., and Garrard, W.L.: Model-follower longitudinal control for automated guideway
transit vehicles. IEEE Trans. Veh. Tech. VT-28,l (1979), pp. 36-45.
124. Rouse, R.J., Jr., and Hoberock, L.L.: Emergency control ofvehicle platoons: control of following-law vehicles. J. Dyn. Sys. Meas. Contr. 98,3 (1976), pp. 239-244.
125. Hoberock, L.L., and Rouse, R.J., Jr.: Emergency control ofvehicle platoons: system operation
and platoon leader control. J. Dyn. Sys. Meas. Contr. 98,3 (1976), pp. 245-251.
126. Zhang, X.: Intelligent driving - PROMETHEUS approaches to longitudinal traffic flow
control. Proc. Vehicle Navigation and Information Systems Conference, Dearborn,
MI, October 1991, pp. 999-1010.
127. Zhang, X., and Benz, T.: Simulation and evaluation of "intelligent cruise control." IVHS J. 1,2
(1993), pp. 181-190.
128. Becker, S., Bork, M., Dorissen, H.T., Geduld, G., Hofmann, O., Naab, K., Nocker, G., Rieth,
P., and Sonntag, J.: Summary of experience with autonomous intelligent cruise control
(AICC) part 1: study objectives and methods. Proc. First World Congress on Applications of Transport Telematics and Intelligent Vehicle-Highway Systems, Paris, November-December 1994, pp. 1828-1 835.
129. Becker, S., Bork, M., Dorissen, H.T., Geduld, G., Hofmann, O., Naab, K., Nocker, G., Rieth,
P., and Sonntag, J.: Summary of experience with autonomous intelligent cruise control
(AICC) part 2: results and conclusions. Proc. First World Congress on Applications
of Transport Telematics and Intelligent Vehicle-Highway Systems, Paris, NovemberDecember 1994, pp. 1836-1843.
130. Sala, G., and Pressi, F.: Strategies for intelligent cruise control (ICC) application. Proc. First
World Congress on Applications of Transport Telematics and Intelligent Vehicle-Highway Systems, Paris, November-December 1994, pp. 2039-2046.
131. Satoh, H., and Taniguchi, I.: A new control method of adaptive cruise control. Proc. First
132. Chien, C.C., and Ioannou, P.: Automatic vehicle-following. Proc. 1992 American Control
Conference, Chicago, IL, June 1992, pp. 1748-1 752.
133. Chien. C.C., Ioannou, P.A., and Lai, M.C.: Entrainment and vehicle following controllers
design for autonomous intelligent vehicles. Proc. 1994 American Control Conference,
Baltimore, June 1994, pp. 6-10.
134. Ioannou, P., and Xu, 2.: Throttle and brake control systems for automatic vehicle following.
IVHS J. 1,4 (1994), pp. 345-377.
135. Fancher, P., Bareket, Z., and Johnson, G.: Predictive analyses of the performance of a headway control system for heavy commercial vehicles. Shen, Z. (Ed.): The Dynamics of
Vehicles on Roads and on Tracks. Proc. 13th IAVSD Symposium, Chengdu, Sichuan,
P.R. China, August 1993. Swets & Zeitlinger, Lisse, 1994, pp. 128-141.
136. Fancher, P., and Bareket, Z.: Evaluating headway control using range versus range-rate relationships. Veh. Sys. Dyh. 23,8 (1994), pp. 575-596.
137. Fancher, P., and Bareket, Z.: The influence of intelligent cruise control systems on traffic flow.
Proc. International Symposium on Advanced Vehicle Control 1994, Tsukuba, Japan,
October, 1994, p. 402-407.
138. Fancher, P.S., Ervin, R.D., Bareket, Z., Johnson, G.E., Trefalt, M., Tiedecke, J., and Hagleitner, W.: Intelligent cruise control: performance studies based upon an operating prototype. Proc. IVHS America 1994 Annual Meeting, Vol. 1, Atlanta, Georgia, April 1994,
pp. 391-399.
139. Foster, E.R.: A constant separation control law for use in automated guideway transit. Proc.
29th IEEE Vehicular Technology Conference, Arlington Heights, IL, March 1979, pp.
358-361.
140. Palmquist, U.: Intelligent cruise control and roadside information. IEEE Micro 13,l (1993),
pp. 20-28.
592
STEVEN E. SHLADOVER
141. Chiu, H.Y., Stupp, G.B., and Brown, S.J.: Vehicle-follower controls for short headway AGT
systems - functional analysis and conceptual designs. Johns Hopkins University Applied
Physics Laboratory Report APL/JHU CP OSI/TPR 035, December 1976, 109 pp.
142. Chiu, H., Stupp, G.B., and Brown, S.J.: Vehicle-followercontrol with variable-gains for shortheadway automated guideway transit systems. J. Dyn. Sys. Meas. Contr. 99,3 (1977), pp.
183-189.
143. Chiu, H.Y.: A state-constrained vehicle-followerapproach to the station-egress problem. IEEE
Trans. Veh. Tech. VT-28,l (1979), pp. 70-79.
144. Miiller, R., and Nocker, G.: Intelligent cruise control with fuzzy logic. Proc. Intelligent Vehicles 92 Symposium, IEEE, Detroit, MI, June-July, 1992, pp. 173-178.
145. Fujita, Y., Akuzawa, K., and Sato, M.: Radar brake system. Proc. International Symposium
on Advanced Vehicle Control 1994, Tsukuba, Japan, October, 1994, pp. 390-395.
146. Yatabe, T., Hirose, T., Tsugawa, S., and Murata, S.: Control of autonomous vehicles with
vehicle-to-vehicle communication. Proc. CCCT 89, JFAC Symposium, Paris, 1989, pp.
553-558.
147. Tsugawa, S., and Murata, S.: Velocity control for vehicle following through vehicle/vehicle
communication. Proc. 22nd International Symposium on Automotive Technology and
Automation, Florence, Italy, May 1990, pp. 343-350.
148. Pue, A.J.: A state-constrained approach to vehicle-follower control for short-headway ACT
systems. Johns Hopkins University Applied Physics Laboratory Report APL/JHU CP
058/TPR 038, August 1977, 128 pp.
149. Pue, A.J.: A state constrained approach to vehicle follower control for short headway automated transit systems. Proc. 1977 Joint Automatic Control Conference, San Francisco, June 1977, Vol. 1, pp. 401-407.
150. Pue, A.J.: A state-constrained approach to vehicle-follower control for short headway automated transit systems. J. Dyn. Sys. Meas. Contr. 100,4 (1978), pp. 291-297.
151. Pue, A.J.: Implementation trade-offs for a short-headway vehicle - follower automated transit
system. IEEE Trans. Veh. Tech. VT-28,l (1979), pp. 46-55.
152. Shladover, S.E.: Longitudinal control of automated guideway transit vehicles within platoons.
J. Dyn. Sys. Meas. Contr. 100,4 (1978), pp. 302-310.
153. Shladover, S.E.: Longitudinal control of automotive vehicles in close-formation platoons. J.
Dyn. Sys. Meas. Contr. 113 (1991), pp. 231-241.
154. Sheikholeslam, S., and Desoer, C.A.: Longitudinal control of a platoon of vehicles. Proc. 1990
American Control Conference, San Diego, May 1990, pp. 291-296.
155. McMahon, D.H., Hedrick, J.K., and Shladover, S.E.: Vehicle modeling and control for automated highway systems. Proc. 1990 American Control Conference, San Diego, May
1990, pp. 297-303.
156. Hedrick, J.K., McMahon, D., Narendran, V., and Swaroop, D.: Longitudinal vehicle controller design for IVHS systems. Proc. 1991 American Control Conference, Boston, MA,
June 1991, pp. 3107-3112.
157. Hedrick, J.K., and Swaroop, D.: Dynamic coupling invehicles under automaticcontrol. Shen, 2.
(Ed.): The Dynamics of Vehicles on Roads and on Tracks. Proc. 13th IAVSD Symposium,
Chengdu, Sichuan, P.R. China, August 1993. Swets& Zeitlinger, Lisse, 1994, pp. 209-220.
158. Sheikholeslam, S., and Desoer, C.A.: Longitudinal control of a platoon of vehicles with no
communication of lead vehicle information. Proc. 1991 American Control Conference,
Boston, MA, June 1991, pp. 31 02-3106.
159. Sheikholeslam, S., and Desoer, C.A.: A system level study of the longitudinal control of a platoon of vehicles. J. Dyn. Sys. Meas. Contr. 114,2 (1992), pp. 286-292.
160. Chang, K.S., Li, W., Devlin, P., Shaikhbahai, A,, Varaiya, P., Hedrick, J.K., McMahon, D.,
Narendran, V., Swaroop, D., and Olds, J.: Experimentation with a vehicle platoon control system. Proc. Vehicle Navigation and Information Systems Conference, Dearborn,
MI, October, 1991, pp. 1117-1124.
161. Choi, S., and Hedrick, J.K.: Comparison of engine models and distance sensors for vehicle
longitudinal control: analysis and test results in PATH. ITS America Annual Meeting,
Washington, D.C., March 1995.
162. Choi, S., and Hedrick, J.K.: Vehicle longitudinal control using an adaptive observer for automated highway systems. Proc. 1995 American Control Conference, Seattle, WA, June
1995.
593
163. Maciuca, D.B., Gerdes, J.C., and Hedrick, J.K.: Automatic braking control for IVHS. Proc.
International Symposium on Advanced Vehicle Control, Tsukuba, Japan, October
1994, pp. 396-401.
164. Ren, W., and Green, D.: Continuous platooning: a new evolutionary and operating concept for
automated highway systems. Proc. 1994 American Control Conference, Baltimore, MD,
June 1994, pp. 21-25.
165. Athans, M., Levine, W., and Levis, A,: On the optimal and suboptimal position and velocity
control of a string of high-speed moving trains. M.I.T. Electronic Systems Laboratory
Report PB 173 640, November 1966, 71 pp.
166. Garrard, W.L., Hand, G.R., and Raemer, R.: Suboptimal feedback control of a string of vehicles moving in a single guideway. Trans. Res. 6 (1972), pp. 197-210.
167. Levis, A,, and Athans, M.: Sampled-data control of high-speed trains. M.I.T. Electronic Systems Laboratory Report ESL-R-339, PB 177 669, January 1968, 50 pp.
168. Powner, E.T., Anderson, J., and Qualtrough, G.: Optimal digital control of cascaded vehicles
in high-speed transportation systems. Trans. Res. 3,1 (1969), pp. 101-1 13.
169. Peppard, L.E., and Gourishankar, V.: An optimal automatic car-following system. IEEE
Trans. Veh. Tech. VT-21,2 (1972), pp. 67-73.
170. Yang, Y.-T., and Tongue, B.H.: A new control approach for platoon operations during vehicle
exitlentry. Veh. Sys. Dyn. (to appear, 1995).
171. Sklar, S.J., Bevans, J.P., and Stein, G.: "Safe-approach" vehicle-followercontrol. IEEE Trans.
Veh. Tech. VT-28,l (1979), pp. 56-62.
172. Caudill, R.J., and Blasnik, S.L.: Model-reference longitudinal control for automated vehicle
systems. ASME J. Dyn. Sys. Meas. Contr. 102, (1980).
173. Caudill, R.J., Di Matteo, P., and Thomas, S.P.: Longitudinal control for automated highway
vehicles. J. Dyn. Sys. Meas. Contr. 104,2 (1982), pp. 180-187.
174. Youcef-Toumi, K., Sasage, Y., Ardini, J., and Huang, S.-Y.: The application of time delay
control to an intelligent cruise control system. Proc. 1992 American Control Conference, Chicago, IL, June 1992, pp. 1743-1747.
175. Godbole, D.N., and Lygeros, J.: Longitudinal control of the lead car of a platoon.: Proc. 1994
American Control Conference, Baltimore, MD, June 1994, pp. 398-402.
176. Grimes, D.M., and Grimes, C.A.: Cradar - an open-loop extended-monopulse automotive
radar. IEEE Trans. Veh. Tech. 38,3 (1989), pp. 123-131.
177. Alvisi, M., Deloof, P., Linss, W., Preti, G., and Rolland, A.: Anticollision radar: state of the
art. Proc. Advanced Telematics in Road Transport, Vol. 11, DRIVE Conference, Brussels, February 1991, pp. 943-961.
178. Stove, A.G., and Chodyniecki, W.: Radar sensor for AICC. Seminar 9 - PROMETHEUS Future Systems, Autotech '93, Institute of Mechanical Engineering, 1993.
179. Kikuchi, H., Ishiyama, M., and Nakajima, T.: Development of laser radar for radar brake system. Proc. International Symposium on Advanced Vehicle Control 1994, Tsukuba,
Japan, October, 1994, pp. 385-389.
180. Swaroop, D., Hedrick, J.K., Chien, C.C., and Ioannou, P.: A comparison of spacing and headway control laws for automatically controlled vehicles. Veh. Sys. Dyn. 23,s (1994), pp.
597-625.
181. Sekine, H., Tamura, K., Asanuma, N., and Sato, M.: A study of electronic cornering speed
control system. Proc. International Symposium on Advanced Vehicle Control 1994, Tsukuba, Japan, October, 1994, pp. 455-460.
182. Butsuen, T., Doi, A., and Sasaki, H.: Development of a collision avoidance system with automatic brake control. Proc. First World Congress on Applications of Transport Telematics and Intelligent Vehicle-Highway Systems, Paris, November-December 1994, pp.
2079-2086.
183.,Naab, K., and Reichart, G.: Driver assistance systems for lateral and longitudinal vehicle guidance - heading control and active cruise support. Proc. International Symposium on
Advanced Vehicle Control 1994, Tsukuba, Japan, October 1994, pp. 449-454.
184. Reichart, G., and Naab, K.: Systems for lateral and longitudinal vehicle guidance. Proc. First
185. Andrews, M., Jacobi, S., Reuter, U., and Salzer, V.: Porche's contribution to safer and more
comfortable driving in the future. Proc. First World Congress on Applications of Trans-
STEVEN E. SHLADOVER
port Telematics and Intelligent Vehicle-Highway Systems, Paris, November-December
1994, pp. 1879-1886.
186. Kehtarnavaz, N., Lee, J.S., and Griswold, N.C.: Vision-based convoy following by recursive
filtering. Proc. 1990 American Control Conference, San Diego, May 1990, pp. 268-273.
187. Sheikholeslam, S., and Desoer, C.A.: Combined longitudinal and lateral control of a platoon
of vehicles. Proc. 1992 American Control Conference, Chicago, IL, June 1992, pp. 17631767.
188. Pham, H., Hedrick, K., and Tomizuka, M.: Autonomous steering and cruise control of automobiles via sliding mode control. Proc. International Symposium on Advanced Vehicle
Control 1994, Tsukuba, Japan, October 1994, pp. 444-448.
189. Pomerleau, D., Thorpe, C., Longer, D., Rosenblatt, J.K., and Sukthankar, R.: AVCS research
at Carnegie Mellon University. Proc. IVHS America 1994 Annual Meeting, Vol. I,
Atlanta, Georgia, April 1994, pp. 257-261.
190. Okuno, A., Kutami, A,, and Fujita, K.: Towards autonomous cruising on highways. Automated Highway/Intelligent Vehicle Systems: Technology and Socioeconomic Aspects.
SAE Publication SP-833, August 1990, pp. 7-16.
191. Morita, T., Takahashi, H., Satonobu, J., Kojima, K., and Morimoto, Y.: An approach to the
intelligent vehicle. Proc. Intelligent Vehicles 93 Symposium, Tokyo, Japan, July 1993, pp.
426-43 1.
192. Hayafune, K., and Yoshida, H.: Control method of autonomous vehicle considering compatibility of riding comfort and vehicle controllability. Automated Highway/Intelligent
Vehicle Systems: Technology and Socioeconomic Aspects. SAE Publication SP-833,
August 1990, pp. 35-40.
193. Ohnishi, K., Komura, J., and Ishibashi, T.: Development of automatic driving system on
rough road - realization of high reliable automatic driving system. Shigematu, T., Hashimoto, Y., and Watanabe, T.: Development of automatic driving system on rough road automatic steering control by fuzzy algorithm. Ooka, N., Tsuboi, T., and Oka, H.:
Development of automatic driving system on rough road - fault tolerant structure for
electronic controller. Proc. Intelligent Vehicles 92 Symposium, IEEE, Detroit, MI,
June-July 1992, pp. 148-165.
194. Suzuki, T., and Tachibana, A.: A fully automated intelligent vehicle system based on computer
vision for automated highway system. Proc. First World Congress on Applications of
Transport Telematics and Intelligent Vehicle-Highway Systems, Paris, NovemberDecember 1994, pp. 1925-1 932.
195. Ulmer, B.: Autonomous automated driving in real traffic. Proc. First World Congress on
Applications of Transport Telematics and Intelligent Vehicle-Highway Systems, Paris,
November-December 1994, pp. 21 18-2125.
196. Hsu, A., Sachs, S., Eskafi, F., and Varaiya, P.: The design of platoon maneuvers for IVHS.
Proc. 1991 American Control Conference, Boston, MA, June 1991, pp. 2545-2550.
197. Varaiya, P.: Smart cars on smart roads: Problems of control. IEEE Trans. on Auto. Contr.
AC-38,2 (1993), pp. 195-207.
198. Lygeros, J. and Godbole, D.N.: An interface between continuous and discrete-event controllers for vehicle automation. Proc. 1994 American Control Conference, Baltimore,
MD, June 1994, pp. 801-805.
199. Rao, B.S.Y., Varaiya, P., and Eskafi, F.: Investigations into achievable capacities and
stream stability with coordinated intelligent vehicles. Trans. Res. Rec. 1408 (1993),
pp. 27-35.
200. Rao, B.S.Y., and Varaiya, P.: Flow benefits of autonomous intelligent cruise control in mixed
manual and automated traffic. Trans. Res. Rec. 1408 (1993), pp. 36-43.
201. Niehaus, A,, and Stengel, R.F.: An expert system for automated highway driving. Proc. 1990
American Control Conference, San Diego, CA, May 1990, pp. 274-280.
202. Niehaus, A., and Stengel, R.F.: Rule-based guidance for vehicle highway driving in the presence of uncertainty. Proc. 1991 American Control Conference, Boston, MA, June
1991, pp. 3119-3124.
203. Niehaus, A., and Stengel, R.F.: Probability-based decision making for automated highway
driving. Proc. Vehicle Navigation and Information Systems Conference, Dearborn,
MI, October, 1991, pp. 1125-1136.
204. Pilutti, T., Raschke, U., and Koren, Y.: Computerized defensive driving rules for highway
205.
206.
207.
208.
209.
210.
595
maneuvers. Proc. 1990 American Control Conference, San Diego, CA, May 1990, pp.
809-81 1.
Karaaslan, U., Varaiya, P., and Walrand, J.: Two proposals to improve freeway traffic flow.
Proc. 1991 American Control Conference, Boston, MA, June 1991, pp. 2539-2544.
Rao, B.S.Y., and Varaiya, P.: Potential benefits of roadside intelligence for flow control in an
IVHS. Proc. 1994 American Control Conference, Baltimore, MD, June 1994, pp. 418422.
Rao, B.S.Y., and Varaiya, P.: Roadside intelligence for flow control in an intelligent vehicle
and highway system. Trans. Res. C: Emerg. Tech. 2C.1 (1994), pp. 49-72.
Ramaswamy, D., Medanic, J.V., Perkins, W.R., and Benekohal, R.: Lane assignment on an
automated highway. Proc. 1994 American Control Conference, Baltimore, MD, June
1994, pp. 413-417.
Peng, H.: Highway-level vehicle control for AHS. IVHS J. (to appear 1995).
Tsao, H.-S. J.: A functional architecture for automated highway traffic planning. Proc. IVHS
America 1994 Annual Meeting, Vol. 2, Atlanta, Georgia, April 1994, pp. 597-605.

International Journal of Vehicle Mechanics and Mobility
Special issue on the 4th International Tyre

Colloquium
Patrick Gruber & Robin S. Sharp
To cite this article: Patrick Gruber & Robin S. Sharp (2016) Special issue on the
4th International Tyre Colloquium, Vehicle System Dynamics, 54:4, 445-447, DOI:
10.1080/00423114.2016.1166599
To link to this article: http://dx.doi.org/10.1080/00423114.2016.1166599
Published online: 04 Apr 2016.
Article views: 124
View Crossmark data

Date: 16 June 2016, At: 03:00
VEHICLE SYSTEM DYNAMICS, 2016

VOL. 54, NO. 4, 445447
http://dx.doi.org/10.1080/00423114.2016.1166599
PREFACE
Special issue on the 4th International Tyre Colloquium

Ever since the first meeting in 1991, organised by Professor Pacejka at the Delft University
of Technology, Tyre Colloquia have provided a platform for researchers and scientists from
industry and academia to discuss and consolidate progress in the understanding and simulation of tyreroadvehicle interactions. In particular, the 1st Tyre Colloquium focused
on the steady-state shear forces and torque generating properties of the tyre. The 2nd
Colloquium, organised by Professors Bhm and Willumeit at the Technical University of
Berlin in 1997, primarily discussed transient tyre characteristics. With the development of
increasingly sophisticated mechatronic automobiles, the topics at the 3rd Tyre Colloquium
(held at the Vienna University of Technology in 2004) ranged from tyre modelling to control aspects of vehicle dynamics. The event was organised by Professors Lugner and Plchl,
who also presented the first results of the international Tyre Model Performance Test. In
2015, the 4th Tyre Colloquium was held at the University of Surrey, Guildford, United
Kingdom. In line with the tradition, the topics focused on current challenges related to
modelling of tyres for vehicle dynamics analysis.
Since the first Colloquium, the capabilities of simulating tyreroadvehicle interactions
have significantly advanced so that nowadays tyre forces and moments can be accurately
modelled within experimentally tested conditions. Yet, many simulation challenges remain
unsolved as tyre mechanics/dynamics is a complicated subject and a full understanding
of the fundamental tyreroad interaction mechanisms is still missing. Today, driven by
the use of high-fidelity simulation tools in the vehicle development process, one of the
main issues is the accurate tyre behaviour prediction outside tested conditions to allow
simulation of all possible facets of running characteristics. Also, considering the increasing connectivity of vehicles, the use of tyres as a real-time sensor offering chassis control
systems direct access to tyre forces and road friction conditions is highly desirable and, at
same time, very challenging.
The 4th Tyre Colloquium focused on todays challenges by discussing a wide range
of topics, including tyre measurements and measurement techniques, application of tyre
models to vehicle dynamics, determination of tyre model parameters, measurement and
modelling of rubber friction, tyreroad contact simulation, and evaluation and verification of tyre models. This special issue includes a small selection of papers presented in
April 2015. The contributions are organised as follows.
The first paper by Gipser [1] provides a clear and structured account of the potential
of predictive capabilities of a tyre model by exploring five different operating scenarios,
such as running on very rough roads. The author demonstrates that physically consistent modelling allows instructive insights into complex tyre behaviour without the need of
modelling every effect, while pointing out the limitations in terms of absolute accuracy due
to the lack of detailed understanding of rubberroad friction. The second contribution by
Wei et al. [2] further investigates the benefits of physically consistent modelling by studying
2016 Informa UK Limited, trading as Taylor & Francis Group
446
PREFACE
and analysing the speed dependency of steering torque characteristics. Using a validated
detailed nonlinear finite-element model, the authors identify turn-slip-related kinematics as the main mechanism for the speed-dependent tyre steering torque. The important
comfort-related aspects of driving noise and vibrations are discussed by Gallrein et al. [3]
by incorporating a structural tyre model into a frequency-based simulation framework.
The authors show that the approach correctly captures tyre behaviour for frequencies of
up to 250 Hz resulting from geometric road excitations. The fourth paper by Schmeitz and
Alirezaei [4] also studies tyre vibrations, but with the view of extracting information on the
tyreroad friction condition. In particular, the resonance peak in the wheel speed signal
excited by road roughness is used to estimate changes in road friction. Results show that
the proposed estimation technique is reasonably accurate. Friction is the main topic of
the fifth contribution by Sharp et al.[5] This study discusses the mechanics of rubber friction by revisiting the classical experiments of Grosch and comparing the measurements
with friction coefficient predictions from Perssons deformation theory. It is shown that
there is value in the current rubber theory but considerable scope for further research on
the mechanics of rubber friction exists. The sixth paper, written by Maniowski [6], closes
the special issue by highlighting the importance of tyres in relation to vehicle handling.
The author explores the influence of temperature-sensitive tyre characteristics on vehicle
behaviour and driving strategy by solving the multi-parametric optimisation problem of
the minimum lap time at a race circuit.
All together, we hope that the reader will find this special issue as interesting as we do
and the new insights and inputs will be useful for researchers and engineers working with
tyres.
We thank the International Technical Committee, the Session Chairs, all participants
and authors who worked hard to help make the 4th Tyre Colloquium a success and facilitated a very interesting programme that allowed a sharing of ideas between experts from
around the world. Also, we are very grateful to the International Association for Vehicle System Dynamics board and, most notably the president Professor Hans True, for the
prompt and full support for organising the Tyre Colloquium at the University of Surrey.
Special thanks go to Professor Manfred Plchl and Jan Prins who were essential in making
this event possible in the first place.
International Technical Committee
Besselink, I., TU Eindhoven, Netherlands
Blanco-Hague, O., Michelin, France
Blundell, M., Coventry University, UK
Gipser, M., Hochschule Esslingen, Germany
Klppel, M., DIK, Germany
Lorenz, B., Forschungszentrum Jlich, Germany
Lugner, P., TU Wien, Austria
Mavros, G., Loughborough, UK
Oertel, Ch., FH Brandenburg, Germany
Pacejka, H. B., Netherlands
Persson, B., Forschungszentrum Jlich, Germany
Plchl, M., TU Wien, Austria
Prins, J., Jaguar Land Rover, UK
VEHICLE SYSTEM DYNAMICS
447
Rauh, J., Daimler AG, Germany

Rill, G., OTH Regensburg, Germany
Schmeitz, A., TNO, Netherlands
Spetler, F., Michelin, France
Wallaschek, J., Leibniz Universitt Hannover, Germany
Zegelaar, P., Ford Motor Company, Germany
Special thanks to the sponsors of the 4th Tyre Colloquium (in alphabetical order)
IPG Automotive GmbH, Jaguar Land Rover, MSC Software, MTS Systems Ltd, SIMPACK.
References
[1] Gipser M. FTire and puzzling tyre physics: teacher, not student. Vehicle Syst Dyn.
2016;54(4):448462. doi:10.1080/00423114.2015.1117116
[2] Wei Y, Oertel Ch, Liu Y, Li X. A theoretical model of speed-dependent steering torque for
rolling tyres. Vehicle Syst Dyn. 2016;54(4):463473. doi:10.1080/00423114.2015.1111391
[3] Gallrein A, Bcker M, Roller M. NVH model of a rotating tire. Vehicle Syst Dyn.
2016;54(4):474491. doi:10.1080/00423114.2016.1158844
[4] Schmeitz AJC, Alirezaei M. Analysis of wheel speed vibrations for road friction classification.
Vehicle Syst Dyn. 2016;54(4):492509. doi:10.1080/00423114.2015.1131307
[5] Sharp RS, Gruber P, Fina E. Circuit racing, track texture, temperature and rubber friction.
[6] Maniowski M. Optimisation of driver actions in RWD race car including tyre thermodynamics.
Patrick Gruber and Robin S. Sharp

Department of Mechanical Engineering Sciences
University of Surrey, Guildford, UK

International Journal of Vehicle Mechanics and Mobility
State of the art survey: active and semi-active

suspension control
H. Eric Tseng & Davor Hrovat
To cite this article: H. Eric Tseng & Davor Hrovat (2015) State of the art survey: active
and semi-active suspension control, Vehicle System Dynamics, 53:7, 1034-1062, DOI:
10.1080/00423114.2015.1037313
Published online: 14 May 2015.
Article views: 1167
View Crossmark data

Date: 16 June 2016, At: 23:57
Vehicle System Dynamics, 2015

Vol. 53, No. 7, 10341062, http://dx.doi.org/10.1080/00423114.2015.1037313
State of the art survey: active and semi-active suspension

control
H. Eric Tseng and Davor Hrovat
Ford Motor Company, 2101 Village Rd, Dearborn, MI 48121, USA
(Received 18 February 2015; accepted 31 March 2015 )
This survey paper aims to provide some insight into the design of suspension control system within
the context of existing literature and share observations on current hardware implementation of active
and semi-active suspension systems. It reviews the performance envelop of active, semi-active, and
passive suspensions with a focus on linear quadratic-based optimisation including a specific example.
The paper further discusses various design aspects including other design techniques, the decoupling
of load and road disturbances, the decoupling of pitch and heave modes, the use of an inerter as
an additional design element, and the application of preview. Various production and near production suspension systems were examined and described according to the features they offer, including
self-levelling, variable damping, variable geometry, and anti-roll damping and stiffness. The lessons
learned from these analytical insights and related hardware implementations are valuable and can be
applied towards future active or semi-active suspension design.
Keywords: active suspension; optimal control; semi-active system; suspension control
1.
Introduction
There are numerous important requirements that a typical automotive suspension must satisfy
[1,2] including
maintaining proper vehicle posture when subject to various inertial and external forces and
moments caused by braking, turning, wind gust, and other events;
providing ride comfort in view of the road roughness inputs which act as a major
disturbance to the vehicle;
securing good road handling and overall vehicle agility;
avoiding excessive suspension stroke or related hard stop/impact.
Additional opportunities exist with the more advanced active and semi-active suspensions
regarding various safety functions [3] and other new exciting functionalities.[4] Although
most current suspensions are of the conventional passive type, this survey will focus on active
suspensions where additional power sources such as pumps and compressors are needed
to accomplish the desired functionality (for a more precise definition of passive and active
*Corresponding author. Email: htseng@ford.com
c 2015 Taylor & Francis
1035
suspensions in terms of the well-established mathematical notion of passive operators, see

[57] and references therein).
While advanced active suspensions offer opportunities for substantial improvements in
ride and handling as well as overall vehicle posture, stability, and added functionality, they
also typically imply higher energy consumption, cost, complexity, packaging, and additional operational requirements. The latter include smooth start-up and shutdown, robust
operations and diagnostics, fault containment and management, as well as the ability to
effectively adapt to different operating conditions including reconfiguration in case of
sensor/actuator/mechatronic system (partial) failure.
As a compromise between passive and active suspension performance, complexity, and
installation and operational costs, one sees more and more semi-active suspensions, which
are marketed under names such as continuously controlled damping, MagneRide, and others. Semi-active suspensions typically include an adjustable damper or shock absorber [8]
where the damping parameter is modulated through a relatively small computer-controlled
actuator. Since it does not require much energy, the semi-active damper can be considered an
essentially passive device.
Complete development of an advanced suspension must consider all of the above requirements. This requires consideration of practically all vehicle degrees of freedom (DOF) within
the bandwidth of main interest, that is, in the case of vehicle main body mass or sprung
mass this would include all 6 DOF (vertical, longitudinal, lateral, pitch, roll, and yaw). However, the main emphasis of the present survey is on ride quality, which is a traditional main
attribute of interest when addressing advanced passenger vehicle suspensions. This would
typically lead to designs using a reduced vehicle sprung mass model with motion limited to
only vertical (or heave), pitch, and roll components.
Moreover, within the above somewhat limited context, the main focus will be on securing superior ride comfort in view of the always present road-induced disturbances caused by
road roughness and irregularities such as potholes and bumps. Thus, we will for the most part
separate the load-induced disturbance discussion such as pitch and roll changes due to (sudden) braking, acceleration, turning, and similar external actions from the ride performance
optimisation analysis. This, of course, does not mean that we are ignoring their importance,
but rather that the engineering task at hand has been modularised or divided into more manageable and insightful subtasks. In the case of road-induced disturbances, lots of insight can
be gained based on simple models which can later be integrated with other subsystems and
subtasks including the aforementioned load-induced dynamic effects.
2.
Optimal suspension control with H2/linear quadratic synthesis/design
As stated in the introduction and as established from many suspension studies,[7,911] the
most significant and insightful conclusions for vehicle suspensions can be observed from a
simple quarter car model with an undamped tyre and a road disturbance described by Gaussian white noise ground velocities or, equivalently,[7] a step in road displacement (or an
impulse in road velocity). Therefore, in this paper, we will focus our discussion on control
analysis and comparison using the quarter-car model.
The performance indices relating ride comfort and handling are often measured by the rootmean-square (RMS) values of sprung mass acceleration and tyre deflection, respectively.[7]
Additionally, the RMS value of suspension displacement can be utilised to enforce the rattle
space (or available suspension displacement) constraint. As a result, the optimisation of the
suspension controller using the quarter-car model often lends itself to a well-defined H2 or
linear quadratic (LQ) optimisation problem.
1036
2.1.
H. Eric Tseng and D. Hrovat
Performance index
The ground vehicle suspension design is influenced by a number of often conflicting factors
and attributes. These factors will be discussed and then combined into an overall performance
index for the H2 optimisation problem.
2.1.1.
Performance index of ride comfort
One of the most popular and fitting ride measures is based on the RMS value of vertical
acceleration of the sprung mass, typically measured or projected at the drivers or passengers
seat locations. In a field study in 1978,[12] the authors involved 78 passengers in two different
vehicles and 18 different road sections to conclude that excellent correlation was found to
exist between the subjective ride ratings and simple RMS acceleration measurement at either
the vehicle floorboard or the passenger/seat interface.
Further refinements of the RMS ride measure are possible through adding the RMS value
of the sprung mass jerk (the derivative of acceleration) to the RMS value of sprung mass
acceleration. Some literature [1315] advocates the inclusion of jerk as an added measure
that amplifies the contribution from high-frequency disturbances, which are important for
NVH (noise, vibration, and harshness).
To account for the frequency dependency of human sensitivity to vibrations and the length
of time of human exposure, a standard has been developed by the International Organization for Standardization as ISO 2631.[16,17] It noted that the region of greatest human
sensitivity to vertical vibration lies between 4 and 8 Hz, which roughly includes the various
resonances of human internal organs. The ISO-based metric can be easily incorporated into
the H2 optimisation formulation.
National Aeronautics and Space Administration (NASA) developed a more comprehensive measure that takes into account interdependency of various modes/direction
of vibrations.[18] A similar measure has been adapted and extended to automotive
applications,[19] considering the interdependence between heave, pitch, roll, and other
factors.
The ride metric of choice will depend on the context of its usage. While the more complex ones can reflect more details and nuances, the simple one can better focus on the
system level and major benefits. This paper will utilise the simple RMS of sprung mass
acceleration.
2.1.2.
Performance index for suspension displacement constraint
In practice, the available suspension displacement or so-called rattle space is limited. One can
include this constraint by adding the suspension displacement in the performance index. With
the attempt to minimise the corresponding cost, the H2 optimisation problem effectively soft
constrains the suspension displacement. Similar to the ride comfort metric, the suspension
displacement-related cost can be reflected on the RMS value of suspension displacement,
especially for random/stochastic road input.
2.1.3.
Performance index for handling/road holding
As there exists a convex relationship between tyre cornering forces and normal forces
(reflected by tyre deflection), the variation of tyre deflection leads to the variation of cornering forces. It is conceivable that a very large variation of tyre deflection may even lead to
1037
a loss of contact with the ground. Therefore, the RMS value of tyre deflection variation can
be an explicit metric for handling characteristics. Note the direct relationship between tyre
deflection and cornering capabilities have been experimentally studied.[7]
Based on the aforementioned discussion, it is no surprise that the most widely used performance index for active suspension study is the combination of the three previously discussed
RMS values. That is, using a weighted combination of RMS acceleration of sprung mass,
RMS suspension displacement, and RMS tyre deflection variation to represent the performance index for optimisation. This performance index lends itself to an H2 norm of the
weighted states and input of the quarter-car model. Interestingly, the deterministic and the
stochastic interpretations of this H2 norm lend themselves to the road disturbance of a step in
road displacement, and that of Gaussian white noise ground velocities, correspondingly. The
equivalence of these two interpretations was described in more detail in [20].
2.2.
H2 optimal control for 2 DOF quarter-car model
The mathematical state space representation of a 2 DOF quarter-car model (Figure 1) can be
found in [9]. Here, we choose the states as tyre deflection, unsprung mass velocity, suspension deflection, and sprung mass velocity (Figure 1). With this representation, we have the
performance index as
PI = E(r1 x21 + r2 x23 + u2 ),
(1)
where x1 and x3 represent the tyre deflection and suspension stroke as illustrated in
Figure 1(a), u represents the sprung mass acceleration (u = U/ms ), and the expectation E
represents steady-state mean-square values. Figure 1(b) shows the corresponding limiting
case when unsprung mass is negligible, and Figure 1(c) illustrates a passive quarter car.
Figure 2 shows a quarter car equipped with various suspension elements including
adjustable spring and damper and different configurations such as springs in parallel to the
actuator and in series to the actuator.
The earlier four-state LQ optimisation was solved numerically,[7,9] and the results of
the corresponding global study for various weighting in the combination of performance
index are shown in Figures 3 and 4. The figures illustrate how the weighting factors serve as
tuning knobs, controlling the trade-off between acceleration, suspension stroke (design constraint), and tyre deflection (handling performance constraint). These plots were generated
with the ratio of sprung-to-unsprung mass set to 10 and the unsprung natural frequency (or
wheel-hop) as 10 Hz. The choice in optimisation with respect to different combinations of
Figure 1. (a) Two DOF quarter-car model, its states x, road velocity w, and actuator input U; (b) one DOF
quarter-car model; and (c) two DOF quarter-car model with passive spring and damper.
1038
Figure 2. (a) Two DOF quarter-car model equipped with actuator, serial spring, and parallel adjustable spring and
damper and (b) the corresponding bond graph model.
weighting in the performance index can be predetermined as the vehicle design suspension
characteristic or adapted dynamically during on-road operation to enable the best driving
experience.
We see from Figures 3 and 4 that reducing r1 , the weighting factor on tyre deflection,
while maintaining r2 , the weighting factor on suspension deflection, results in increased tyre
deflection and decreased sprung mass acceleration, that is, better ride at the expense of somewhat worse handling. On the other hand, keeping r1 constant while reducing r2 leads to
increased suspension deflection and decreased sprung mass acceleration. In the extreme case
of assigning r2 to zero, the problem becomes a degenerative one where extremely large suspension displacement would result. The corresponding optimal force of this degenerated case
would not attempt to support the static weight of the vehicle since there is no cost associated
with suspension stroke. As a rule of thumb, choosing r1 /r2 to about 10 leads to reasonable
results.[7] The shaded area in Figures 3 and 4 illustrates where most optimal designs for
automotive applications will reside due to the above-mentioned practical considerations. An
important observation is that no suspension control design, linear or nonlinear, can exhibit
behaviour in the area falling below the r1 near zero line (represented by r1 = 102 ) or the r2
near zero line (represented by r2 = 103 ).
Similar solutions can be obtained using linear matrix inequality (LMI)-based
approaches.[2123] A desirable feature of the LMI-type controllers is that they can incorporate the slowly time-varying parameters, such as changes in vehicle sprung mass. Related
methods here include the linear parameter-varying techniques [22,24] that explicitly include
possible slowly varying system parameters, such as road roughness and vehicle speed
variations with results similar to gain scheduling investigated in [25].
Various observations about the suspension performance limitation from different
approaches as well as limited experimental work [7] have affirmed the performance limits
observed in Figures 3 and 4. For example, any improvement in one area (e.g. ride) would be
reflected through corresponding loss of performance in other complementary areas (e.g. handling in wheel-hop mode and/or rattle space requirement). Essentially this is a consequence
of the fact that a single actuator is required to fulfil two conflicting requirements. This is also
reflected through the invariant properties reported in [26].
1039
Figure 3. Normalised ride comfort versus suspension stroke (A as road roughness coefficient and V as vehicle
speed).
2.3.
Illustrative example
To illustrate the performance in frequency response as well as where it resides on the global
carpet plot of Figures 3 and 4, a practical example of H2 optimal design active suspension
is described in the following.
Assume that for a particular vehicle = 10 and us = 2 10 rad/s. The road is described
by a road roughness coefficient A = 1.6 105 ft (4.9 106 m) which corresponds to a
medium quality road. The vehicle is traversing this road at a speed of V = 80 ft/s (88.5
km/h). It is reasonable and desirable to design a suspension such that the tyre deflection from
equilibrium under this road excitation would be maintained within 1 inch (2.54 cm) 99.7%
of the time. This would translate to a requirement on the normalised tyre deflection, x1 , to
RMS
be maintained within 0.3 s1/2 . Note that we use the accent to denote the normalised
value with respect to the road velocity excitation, for example, x1 := x1,rms / 2 AV , where
A is the road roughness coefficient and V is the vehicle travelling speed.[27]
From Figure 4 and a normalised RMS tyre deflection value of 0.3 (s1/2 ) in this illustrative
example, we see the minimum possible normalised RMS acceleration u is 10 (s3/2 ). If we
accept the normalised value of 10.9 (which reflects to a 0.03 g sprung mass acceleration RMS,
1040
Figure 4. Normalised ride comfort versus handling.
or the sprung mass acceleration under 0.1 g 99.7% of the time), we can arrive at this optimal
control law by picking the tuning factors r1 = 1100 and r2 = 100. This is represented as
design point A in Figures 3 and 4. Using these values for r1 and r2 in Figure 3 reveals a
normalised RMS secondary suspension deflection x3,rms = 0.605 s1/2 which ensures that the
suspension deflection will remain within 2 inch (5.08 cm) of the static value 99.7% of
the time. With r1 = 1000 and r2 = 100, the optimal control gains for full state feedback are
k 1 = 6.084, k 2 = 0.548, k 3 = 10.0, and k 4 = 4.438 for tyre deflection x1 , unsprung
mass velocity x2 , suspension deflection x3 , and sprung mass velocity x4 , respectively. The
closed-loop eigenvalues are e1,2 = 2.20 j2.26 and e3,4 = 2.75 j62.9. The first set
of oscillatory eigenvalues corresponds to the vehicle heave mode with a natural frequency
of 0.5 Hz and damping ratio of 0.7. The second set corresponds to the wheel-hop mode
with a resonant frequency of 10 Hz and relatively small damping ratio of 4.4%. Whether or
not this small amount of damping is of consequence will depend on the particular hardware
implementation, which may include a combination of active and passive means to improve
closed-loop robustness. Also in practice, an excessively soft suspension setting may require
load levelling to keep the suspension static or steady-state deflection in the neighbourhood of
zero or some nominal value.
1041
Figure 5. Normalised ride comfort versus handling passive and active suspensions.
At this stage, it is instructive to compare the above active suspension results with a corresponding passive case with the structure shown in Figure 1(c). For this purpose consider
Figure 5 which displays normalised RMS acceleration versus tyre deflection for active and
passive suspensions. For simplicity, the optimal active suspension performance is shown via
the limiting curves r1 0 and r2 0 only, which are representative of the main trends.
Here, the design points with lower accelerations and higher tyre deflections are characterised
by softer, lower frequency body modes and less damped wheel-hop modes.
The added curves in Figure 5 correspond to the passive performance trade-offs with sprung
mass mode resonant frequencies f2 at 1 or 1.5 Hz (which corresponds to different passive
spring stiffness in Figure 1(c)), and damping ratios varying between 0.02 and 1 (which
corresponds to varying passive damping in Figure 1(c)). A typical passive suspension setting
is shown as design point P1 , where f 2 = 1 Hz and = 0.3. As can be seen from Figure 5,
this amount of damping is near optimal in terms of the ride and handling compromise for the
passive case. These values are within the range typically found in most of todays vehicles
1042
which have evolved through many iterations primarily based on intuition and decades of
experience.
The performance of an optimal active suspension with the same amount of wheel-hop
as the optimal passive is shown in Figure 5 as design point A1 . Comparing design points
P1 and A1 , it can be seen that in this case the active suspension results in only 11% lower
RMS acceleration levels with respect to P1 . However, it can also be seen that for some other
road/speed operating conditions (to the right of A1 ), there is a potential for substantial ride
improvement with active suspensions compared with their passive counterparts. Indeed, as
shown in Figure 5, for passive suspensions almost any deviation in operating conditions from
P1 will result in degradation of performance. On the other hand, for active suspensions, either
handling or ride can be improved by choosing the tuning parameters r1 and r2 such that the
resulting operation settles either to the left or to the right of A1 , respectively. For example, if
we pick a different active suspension design point A (Figure 5) by relaxing the tyre deflection
constraints, we see that the RMS acceleration at point A is as much as 67% smaller than that
at Point P1 .
Based on the above results and observations, it can be concluded that the full advantage
of active suspensions stems from possible adaptive tuning (or gain scheduling) of controller
parameters, depending on the driving condition.[7,25] For example, if the steering wheel
position or lateral acceleration sensors (possibly augmented with other road information)
indicate operation on a straight section of a road where handling is less critical, then it may
be possible to relax the wheel-hop constraint. Similarly, by knowing which sections of the
road the vehicle is or will be travelling on and their corresponding road roughness, the tyre
deflection constraints can be enforced differently, allowing more substantial improvement in
RMS acceleration. Note that the exact amount of wheel-hop that could be tolerated under
different driving conditions can be predetermined through appropriate vehicle tests.
The frequency responses of the three metrics with respect to road disturbance velocity offer
insightful information between the trade-offs in suspension design. Different attributes of the
performance at three operating points of interest (A, A1 , and P1 ) are illustrated in Figure 6
in their corresponding Bode plots. From the sprung mass acceleration frequency plot, it can
be seen that the main difference between the active suspension A1 and corresponding passive
counterpart P1 occurs at the sprung mass mode around 1 Hz, where the active suspension
brings much more damping through the damping on sprung mass velocity, the so-called
skyhook damping. The active suspension A results in further significant reduction of the
acceleration levels (except at the invariant point at the wheel-hop frequency [26]), but at
the same time it deteriorates handling and rattle space performance due to a large, poorly
damped resonant peak around the wheel-hop frequency. Note that the invariant point arises
from the structure of the suspension and results in the performance limit of sprung mass
acceleration at the wheel-hop frequency regardless of control design, assuming negligible
tyre damping. The reduction of acceleration at other frequency regions causing the rise of
suspension and tyre deflection can also be interpreted as the waterbed effect, a term coined
in H infinity control analysis, that is, the improvement in one area (e.g. ride with respect to
sprung mass frequency) of the frequency domain would be reflected through corresponding
loss of performance in other complementary areas (e.g. handling in tyre-hop mode and/or
rattle space requirement).
The suspension deflection Bode plot of Figure 6(b) indicates that, unlike their passive
counterparts, the LQ optimal suspensions can result in a non-zero gain in very low frequency
between the rattle space and road velocity input.[7] However, in practice this may not be
detrimental since it is usually a standard practice to use appropriate signal processes, for
example, high-pass filters to eliminate the very low-frequency components of the ground
input signals resulting from hills and extended road grade.
1043
Figure 6. (a) Tyre deflection Bode plot of points on Figure 5 and point A with dynamic absorber, (b) suspension
Bode plot of points on Figure 5 and point A with dynamic absorber; and (c) acceleration Bode plot of points on
Figure 5 and point A with dynamic absorber.
Moreover, Figure 6 shows the Bode plots of an active suspension control (designed with
the same weighting used in operating point A) combined with a dynamic vibration absorber.
We see that the sprung mass acceleration can be further improved in the wheel-hop frequency
since the introduction of the dynamic absorber (which interacts with only the unsprung mass)
presents a new structure.
2.4.
Extension to semi-active optimal case
In semi-active suspension systems, the suspension force can be modulated through a very
small amount of supplied energy. In this case the suspension force cannot track an arbitrary desired force from the unconstrained LQ optimisation-derived control law. Rather, the
problem becomes a constrained LQ optimisation. As a compromise, the semi-active control
design typically follows its unconstrained active counterpart when it can, and operates along
the passivity envelope when it cannot. For example, the damping force is adjusted to follow
the desired suspension force derived from the optimal control law, and set to zero when a negative damping force is required. This control is therefore commonly referred to as clipped
optimal. This may result in 20% lower performance in ride comfort when compared to its
active suspension counterpart.[11]
1044
The optimal control law for the semi-active system has been posed as a constrained LQ
optimisation and solved numerically in [28,29], involving the iterative solution of a timevarying force constraint. A specific example [29] shows that a 10% advantage with respect
to clipped optimal can be achieved. However, it also finds that the amount of improvement
depends on driven scenarios and is usually very limited. A later work [30] leveraged the
explicit hybrid MPC to confirm analytically the previously numerical finding that clipped
optimal is not the optimal control for semi-active suspensions in general.
2.5.
Extension to preview optimal control
There are numerous investigations of the potential performance of the suspension if the road
disturbance profile can be known in advance.[3135] The solution consists of state feedback
terms that have the same gains as the non-preview case, and feedforward terms that consider
the road disturbance of the upcoming known duration.
It is worth noting that most of the papers place the origin of time (i.e. start of preview
time for future disturbances) at the quarter-car location [31,35,36] or at the location of the
front wheels.[32,37,38] In [34], this origin is simply moved to the location of preview,
thus transforming the preview problem into an equivalent delay-time problem, where an
extensive literature base already exists.[39] This approach is also used in [40] where the
authors developed a method for direct and efficient calculation of the related performance
index.
Comprehensive studies of the benefit of preview compared to its non-preview counterpart
have been demonstrated for full-bandwidth active suspensions [34] and limited bandwidth
ones.[41,42] The benefit for full-bandwidth active suspensions was illustrated on a plot similar to Figures 3 and 4 in [34], relating to its non-preview counterpart. And the benefit for
limited bandwidth active suspensions in ride comfort and suspension stroke improvement
was shown in [42].
3.
Special consideration of handling, loading, and suspension travel limit
Although not always explicitly mentioned, most of the studies with traditional simple ride
models (e.g. quarter-car, half-car, and full-car) focus on road roughness disturbances only,
neglecting external (inertia) forces and loads due to, for example, braking and turning. The
control damping for inertia load-induced excitations can be included later in a separate and
independent design procedure. In practice, this is accomplished using longitudinal and lateral
acceleration measurements as feedforward signals [43] indicating braking or turning actions.
This typically leads to appropriate stiffening of the active or semi-active suspension to prevent or counteract excessive pitch or roll. This stiffening action may reduce the ride quality,
but in practice this may be of secondary importance, especially during hard braking and/or
handling manoeuvres.
For the case when complete decoupling of the above road and load disturbance effects is
desired, Smith,[44] Smith and Wang,[45] and Wang and Smith [46] present an interesting
approach based on Youla parameterisation and LFT (linear fractional transformation). The
resulting controller can be designed independently first for the road-induced disturbance
and then for the load counterpart. For example, the suspension can be made very stiff for
load inputs in order to hold a desired attitude, while at the same time it could be made very
soft for the ground inputs (of course, the latter has to be within some practical limits since
any prolonged ramp-like ground input could lead to suspension bottoming and undesirable
1045
jerks and possibly even structural damage). This is somewhat similar to the above-mentioned
sequential design approach used in practice with the exception of total decoupling. In this
context, the practical design uses longitudinal and lateral acceleration signals (in addition
to other heave-related measurements) as a feedforward signal or proxy for gas/brake pedal
and/or steering wheel applications.
A comprehensive and direct approach to simultaneous inclusion of road and load disturbances has been recently presented in [47] based on the above more conventional LQ
stochastic optimisation. In addition to the usual road disturbance, which is assumed unmeasured, Brezas and Smith [47] also introduce the load disturbance but as a measured, constant
disturbance. An advantage of this approach is that it directly addresses optimisation of the
main attribute of interest, that is, ride comfort, which is introduced through an appropriately
formulated LQ performance index.
In the case of longitudinal dynamics, further extensions of this approach are possible by
introducing the force disturbance through a delay, which effectively presents a load preview
from the standpoint of the suspension controller similar to the preview modelling used in
[34]. This could be further refined by using a (simple) model of vehicle power train (brake)
paths, from gas (brake) pedal to actual wheel torque production. Similar comments apply for
load prediction due to steering inputs where ultimately some appropriate models of vehicle
handling can be used too.
A simple approach to load disturbance containment is through load levelling, that is,
applying an integral control to regulate the rattle space between sprung and unsprung
masses.[4851] In particular, Hrovat and Hubbard [48] use jerk-optimal LQ control to end
up with an effective and systematic way of obtaining relatively fast load levelling with
some mostly minor sacrifice in sprung mass acceleration performance. It should be
pointed out that in practice a slow-acting load levelling is often used to centre the suspension against slow-changing loads. Davis and Thompson [49] use both integral and derivative
controls in a ride quality- and handling capability-focused 7 DOF full-car model to regulate
as well as maximise usable suspension deflection as the derivative action helps to decrease
overshoot.
As the suspension stroke is limited, practical design of all suspensions may introduce
additional feedback elements to stiffen up the suspensions as the rattle space approaches
its travel limits. In addition, actuator dynamics may contain delays, nonlinearities (which is
inherent in the semi-active case), or lags. Various control designs have been developed to
address these issues including backstepping design,[52] adaptive robust control,[53] learning automata,[54,55] and Pareto optimisation.[56] In [52], a nonlinear filter whose effective
bandwidth depends on the magnitude of the suspension travel was introduced to facilitate smooth shifting of the control objective between ride comfort and suspension stroke
utilisation. In [53], a two-loop controller with adaptive robust control was introduced to
ensure that the nonlinear actuator delivers the force requested from the LQ top (outer) loop
design. Gordon et al. [54] and Howell et al. [55] used an approach combining concepts from
stochastic optimal control with those of learning automata. And Valasek and Kortum [56]
used the Pareto optimisation procedure to determine the state-dependent feedback gains in a
predefined control structure.
It is worth noting that Gordon and Best [57] included non-quadratic terms for the optimisation of semi-active suspensions in their optimisation and observed the performance with
the conclusion that the standard clipped optimal technique is close to the one resulted
from a complex optimisation. In [58], the optimisation of semi-active suspension with preview is numerically optimised and it was observed that only small improvement can be
achieved compared to the simple clipped optimal preview control. These conclusions are
corroborated in [30] where analytical explicit model predictive control was used.
1046
4.
Other approaches and considerations
In addition to the above LQ optimisation-based approaches that are the main focus of the
present survey, there are many other possible approaches to advanced active and semi-active
suspension design and analysis. These include Youla parameterisation and LFT,[4446] pole
placement,[59] LMI,[60] H2/H-infinity, and H-infinity.[6165]
Furthermore, multi-objective optimisation (symbolically or numerically) has been widely
used to systematically manage nonlinearities and constraints.[6668] It is particularly
prevalent for semi-active suspension applications.[69,70] As the suspension performance
index often includes conflicting terms, the solution of a multi-objective optimisation finds
the best trade-off among the various predefined control terms. This is also known as Pareto
optimisation used in systematic design procedures (note that a similar result for the linear
system was obtained using the previously mentioned global optimisation in Section 2 with
varying weights).
While ride comfort is the main focus for passenger vehicle suspensions, minimising road
damage is important for truck and heavy vehicle suspensions.[7174] An approach that pays
particular attention to tyre force variation is an interesting extension of the well-known
skyhook concept,[8] proposed in [75] as the so-called ground-hook control. Specifically, it
amounts to augmenting the (semi-)active actuator desired force calculation with an additional, virtual damping term proportional to the relative velocity between the unsprung mass
and ground (Figure 7), emulating a fictitious damper between the unsprung mass and road.
The typical objective of such a suspension optimisation was to minimise the integral square
of the tyre force [76,77] by using a simple quarter-car vehicle model for the control synthesis, which is similar to previously mentioned LQ-based RMS tyre deflection minimisation
except that the latter would typically be part of a performance index that includes additional
terms due to ride comfort and rattle space (design) constraint. The integral square sprung
mass acceleration term as a measure of ride comfort was later included in [78,79] where
the ground hook and other coefficients were determined from numerical optimisation with
specified cost function. Note that the above optimisation for the case of semi-active actuator
amounted to essentially nonlinear system parameter optimisation since the extended groundhook optimisation structure was fixed a priori.[76,77] Since the main nonlinearity is in the
semi-active actuator, state dependence of the related parameters was also introduced in the
process of optimisation.[56]
The ground-hook concept was found useful when minimising road damage [56,77] especially due to heavy trucks where the damage to pavement can be substantial and up to 10,000
times greater than due to passenger cars.[78] This leads to so-called road-friendly suspension,
Figure 7. (a) Skyhook, (b) ground hook, and (c) combination of skyhook and ground hook.
1047
which was further extended to bridge-friendly [7981] and ESC,[82] that is, Electronic Stability Control-friendly, applications. The concept was further tested through experimental
vehicle implementation [56,77] using a SKODA-LIAZ prototype truck.
The main emphasis in Section 2 was on advanced active and semi-active suspensions
using simplified (minimal component count) system models. In practice, however, any implementation of these advanced concepts will also include additional conventional, passive
components. For example, an active actuator may be positioned in parallel with a spring
for static support and a damper to dissipate energy without an active element. Thus, one must
eventually consider designing the whole suspension system consisting of both active and
passive elements.
In addition, as is typically the practice when implementing the LQ or any other linear
algorithm, one should first attempt (static) linearisation of any possible nonlinearities in actuators, plants, and/or sensors. The combined controller-actuator subsystem should then be
further verified and as necessary refined through additional simulations based on a more
detailed performance-oriented model. The latter should include a detailed model of actuator dynamics, which is often the most challenging and most important component of an
advanced, active, or semi-active vehicle suspension system design.
A reasonable starting point in an active or advanced suspension system design would be
a consideration of the passive system alone in order to establish the best passive benchmark
against which any advanced mixed passiveactive system could then be compared. At this
stage there is a need to more precisely define passivity so that the whole passive suspension
optimisation problem can in turn be properly defined. One of the first attempts at this was in
[28] where the notions of passive operators from mathematics and electrical circuits are used.
In addition, Hrovat [28] also posed the corresponding optimal passive suspension problem.
However, at the time of its writing, much of the necessary mathematical Computer Aided
Engineering (CAE) tools and computer optimisation capability were missing. This class of
problems were effectively treated more than a quarter-century later [83] using LMIs and
other modern CAE tools. In [83] problems with both H2 and Hinf optimisation cost are
considered for the case of a quarter-car model with both ground and load inputs included.
This approach could likely be further extended to hybrid configurations where both passive
and active suspensions are combined.
Once the optimal passive transfer function is determined, there remains the problem of synthesising or actually implementing it via standard passive elements such as springs, dampers,
and inertias. In the process, there is a need for an inertia element called inerter in [84]
that acts on a relative velocity rather than the customary absolute velocity of relevant
masses or inertias. Perhaps the easiest way to see this is via bond graphs where one could
imagine a hydraulic pistoncylinder combination between sprung and unsprung masses [85]
(inerter-like structure was described in its third edition with more examples exhibited in its
later editions). Another example can be found in modelling hydraulic mounts.[86,87] An
effective and practical mechanical implementation of an inerter was shown in [83,84] consisting of multiple gear sets with large gear ratios resulting in a large reflected inertia. The
latter facilitates relatively small overall weight for the device, which is an important practical
consideration. In addition to possible suspension applications, the device was also used to
mitigate steering instabilities of motorcycles.[88]
Another more constrained approach to optimisation was considered in [89] where the
underlying passive structure was first fixed and then the optimisation was performed for each
free design parameter and compared with the best tuning of the conventional passive suspension structure consisting of a parallel spring and damper. The results indicate around 10%
and more improvements in the comfort RMS acceleration metric, although it is not clear if
this included the same level of other constraints (tyre and rattle space).
1048
Using the above-mentioned definition of passivity it is now possible to ask the following question. For a given optimal suspension control law can we determine if it could be
implemented through a passive device? If the answer is positive this would have significant
practical implications since such a device could then be realised using only passive components that are typically less expensive, less complex, and require no external power sources
and consumption to operate. It was first shown in [6,28] that for the quarter-car LQ optimal control the resulting control law is necessarily active. A more comprehensive study was
subsequently undertaken in [5] including additional control laws applied to half- and full-car
cases.
In addition, Smith [44] and Smith and Walker [5] consider some other interesting aspects of
active and passive suspensions. These include the derivation of a complete set of constraints
for the road disturbance transfer functions for typical choices of measured outputs as well
as performance limitations and constraints for active and passive suspensions for quarter-,
half-, and full-car models. This complements the prior well-known results on the invariance
of certain transfer functions at special frequency points.[26]
While the quarter-car model largely describes the vehicle response to the road disturbance,
the heave and pitch experienced at various observed points along the vehicle longitudinal axis
can vary due to the multi-input disturbance (coming through both front and rear wheels). For
passive vehicles, due to the fact that the excitation at the rear wheel is a time-delayed version
of the front wheel, there is a wheelbase filtering effect [90] that changes the transfer function
to heave and/or pitch at particular frequencies as a function of vehicle speed and wheel base.
This effect is more perceptible for trucks with a long wheel base. The quarter-car suspension
design can be expanded to full car. One attractive approach to address the full-car and half-car
suspension design and leverage the extensive knowledge from the quarter-car study is to first
apply control design decoupling modes. The importance of decoupling between sprung mass
pitch and heave mode as well as between the equivalent front and rear quarter-car models
is described in [7,91]. In [91], an active suspension was designed to decouple sprung mass
heave and pitch mode through zeroing the coupling terms between pitch and heave relative
(suspension) deflections, and similarly those between pitch and heave relative (suspension)
velocities. This structurally decoupled active suspension minimises body heave motion to
road pitch disturbance, and similarly body pitch to road heave, to a secondary level. It was
exhibited [91] that a superior performance can be achieved by simply adding the skyhook
(or absolute) damping to the decoupled active suspension design. This approach is further
verified in [92] through mathematical manipulation and state space transformation where the
heave and pitch modes and corresponding disturbances are decoupled.
Important aspects of state estimation are addressed in [93,94]. Analytical techniques for
fault detection and facilitating safe start-up and shutdown of active system are described
in [95]. Some recent attempts to introduce risk sensitivity into the LQ optimal suspension
problem were addressed in [47]. The approach has some practical appeal since it facilitates
tuning the balance between performance and associated risk due to plant uncertainties, modelling limitations, and possible unmodelled dynamics, thus providing system designers with
additional flexibility through an added degree of tuning freedom.
5.
5.1.
Production and near production features and systems

Self-levelling feature
Various self-levelling systems have been introduced to the market including the Electronic
Air Suspension systems developed by Citroens hydropneumatic systems [96] and Ford
1049
Motor Companys Electronically Controlled Air Suspension on Lincoln,[97] allowing these

vehicles to maintain proper ride height and suspension stiffness over a wider range of vehicle loading. This feature adjusts the vehicle ride height, usually at low frequencies, in order
to balance among (1) soft and comfortable ride stiffness from the passive spring, (2) proper
vehicle attitude/stance, and (3) ample rattle space for anticipated or unknown road disturbance ahead. For example, the system in the 1984 Lincoln Continental allows its spring to be
3238% softer than a corresponding steel spring suspension while maintaining vehicle stance
and suspension travel.[97] In addition, this active suspension feature can enable the lowering of the vehicle at highway speeds to improve aerodynamics. This feature is implemented
in the Lincoln Mark VIII [97] and most recently the Tesla Motor Model S,[98,99] providing improved aerodynamics, more consistent handling, better fuel economy, and extended
range.
Load levelling suspensions are often constructed using pneumatic springs whose equilibrium length is adjusted by changing the volume of air in the spring. In the Lincoln
Continental, height sensors are used to trigger the adjustment of the air volume, and signals from the ignition, doors, and brake systems are also used to ensure levelling only at
appropriate times.
5.2.
Variable suspension damping (semi-active suspensions) feature
Production semi-active suspension systems are generally constructed using an adjustable

damper in parallel with the primary suspension spring. These are typically constructed from
pneumatic and/or hydraulic piston/cylinder combinations with electromechanical control of
an orifice. Actuator bandwidth is primarily determined by the reaction time of the controlling valve and associated pressure/force production dynamics. Semi-active suspension
systems have been introduced to the market since the 1980s Toyota Soarers TEMS (Toyota
Electronic Modulated Suspension).[100] A more recent system can be found in Lincolns
Continuous Controlled Damping system introduced in 2006. The Lincoln system uses a
suite of sensors that constantly monitor suspension motion, body movement, steering, and
braking inputs and adjusts the suspension in milliseconds, helping keep the car smoothly on
track. Specifically, it monitors up to 46 inputs and reacts on average within 20 milliseconds
[101,102] to reduce roll, pitch, and heave motions, to enhance driving comfort and dynamics,
and to isolate road harshness.
Another implementation of adjustable damping is through magneto-rheological (MR) fluids. MR fluid characteristics can be changed electronically, allowing the force across the
actuator to change quickly. This method benefits from faster response time, although limited
fluid life can contribute to service concerns. One MR damper application is found in the 2002
Cadillac Seville STS and the 2003 Chevrolet Corvette whose MR fluid is co-developed by
Delphi and Lord Corporation. It is composed of suspended iron particles in a base fluid of
synthetic hydrocarbon. Its continuously variable damping is adjusted in milliseconds. The
fast response time of the MR damper enables closed-loop vehicle stability control through
the control of lateral and longitudinal load transfer characteristics of the suspension during
transient movements.[103,104]
Another method to increase the damping rate adjustment frequency when employing
limited bandwidth orifice control valves is to employ the resistance control semi-active damping method [105] demonstrated experimentally in [106] where a dual-acting damper with
two control valves is used. Each valve sets the damping rate in one direction (jounce or
rebound).
1050
5.3.
Variable suspension geometry (low-power low-bandwidth active suspensions)

feature
A variable geometry active suspension adjusts the ratio of wheel movement to the deflection
of the suspension spring in real time. By changing the leverage of the passive suspension
spring with wheel motion, it essentially controls the wheel rate or effective spring stiffness.
Various systems and hardware configurations that provide variable suspension geometry have been proposed in the literature [107110] including the Delft Active Suspension
concept that was implemented as a prototype vehicle and demonstrated experimentally.[109]
Ideally, one would like to adjust the variable suspension geometry without requiring much
power. One way to do this is to adjust the suspension leverage by changing the suspension
spring mounting point which requires only actuation perpendicular to the (vertical) suspension load or vehicle weight (Figure 8).[108] Ideally, with the perpendicular configuration, the
mechanism would require very low power and low energy. In practice, however, the precisely
perpendicular arrangement would be compromised by suspension motion and deflection.
There may be additional trade-offs in mechanical design forcing some deviation from the
ideal geometry. The trade-offs may include vehicle stance/self-levelling, range of mechanism motion (and resulting range of leverage ratio), packaging, effect of jounce/rebound
motion, and effects on wheel turning. A system design study with the trade-offs in mind
is discussed in [107].
Another form of leverage adjustment concept (Figure 9), the previously mentioned Delft
Active Suspension, is described in [107,111], and its practical realisation with a cone mechanism is illustrated in [112,113]. This mechanism connects the spring to the car body on one
end and to a rotatable crank on the other end. The crank is joint-connected to the suspension/wheel control arm and can be rotated at the joint around the base of the imaginary cone.
The cone mechanism serves two purposes: (1) the length of the spring remains the same as
the crank rotates and (2) the ratio of movement between the wheel/tyre control arm and the
crank changes as the crank rotates (Figure 10).
Due to suspension deflections and motions when experiencing road disturbances, the alignment of the ideal position of the cone mechanism will be affected in practice. In particular,
the base plane of the cone mechanism, that is, the plane of the crank motion, will rotate.
Nevertheless, this is a promising technology since it could provide some active suspension
characteristics with only a semi-active type of actuator.
5.4.
Roll control (semi-active or active) feature
A number of automobile manufacturers have been introducing active and semi-active suspensions that focus on roll control to improve ride and handling with moderate cost and
Figure 8. Illustration of a variable geometry suspension concept.
1051
Figure 9. Illustration of delft active suspension concept.
Figure 10. Illustration of delft active suspension realisation Fv cos( ) b = FD a.
complexity. It is well known that good ride comfort can be achieved with softly tuned passive or semi-active suspensions combined with soft roll bar settings. However, this will result
in large pitch and roll motions during more aggressive braking and cornering. The active
(or semi-active) roll control is then used to satisfy these conflicting constraints. In practice,
the concept is typically realised by inserting a small hydraulic jack and a gas spring accumulator with a control solenoid, all placed in series with the anti-roll bar. The anti-roll bar
can be as stiff as in street race cars, without negatively impacting straight-line ride since the
solenoid connecting the hydraulic jack and the accumulator is then kept open, thus lowering
the effective stiffness of the roll bar/accumulator combination.
The performance of anti-roll/roll control devices varies as it may be achieved by a mechanical regulator that depends on suspension motion [114,115] or by an electronic controller
that monitors additional non-suspension-related vehicle signals. It can be actuated with a
semi-active device that manipulates hydraulic valves or the like, or with an additional active
1052
device such as a high-capacity pump. An idea similar to active tilt control has also been
explored in [116,117].
Citroen introduced hydropneumatic suspensions in 1955 with its model DS. The hydropneumatic spring-absorber system incorporates the benefits of both the compressibility of
gas (nitrogen) and responsiveness of (incompressible) hydraulic fluid. As the load increases,
it compresses the suspension pistons and causes the gas volume to decrease and pressure to
rise; the gas spring becomes stiffer but keeps the sprung mass frequency roughly the same.
As the fluid passes through a two-way valve, damping is also provided. In addition to its selflevelling feature that regulates ride height and resonance frequency (through stiffer spring
with increased load), the Citroens hydropneumatic system can be applied towards anti-roll
or roll control by manipulating the flow of hydraulic fluid between left and right corners.
Various improvements have been made in Citroens hydropneumatic suspensions. Most
notable are the changes from mechanical regulation to electronic control, the improved capability of the pump, and the associated increase in actuation bandwidth. For the roll control
feature in particular, there have been several incarnations ranging from semi-active (Hydractive I and II) to active (Activa). Citroens semi-active roll control (Hydractive) limits the roll
angle to about 3 in soft mode or 2.5 in hard mode for a 0.6 g constant turn, while its active
roll control (Activa) can eliminate roll angle completely due to the active and electronic
manipulation of hydraulic fluid.[96]
Roll control can be achieved with the help of a hydraulic rotary actuator inserted around
the middle of the stabiliser bar, as stated above, or it can be implemented through a rotary
actuator fitted with the anti-roll bar as in the BMW Dynamic Drive concept.[118] The BMW
system monitors primarily the steering angle and lateral acceleration, among other signals,
and distributes the front and rear axle anti-roll torques. It controls the front and rear actuators
by means of two electronically regulated pressure control valves to minimise the roll motion
in corners and road input roll disturbances, thus improving response to vehicle steering for
vehicle dynamics handling and agility. Note that the roll angle will gradually increase at high
lateral acceleration as the vehicle approaches its stability limit.
It is understood that the benefits of active roll control in overall vehicle stability and active
safety can be further explored through coordinated action with other systems such as active
brakes, active steering, and electronically controlled all-wheel drive. This can be particularly
useful in emergency manoeuvres such as a sudden need for vehicle stabilisation and control
during evasive actions on slippery road surfaces.
5.5.
Active suspensions with full features
In general, active suspensions can be categorised into narrow bandwidth and broad bandwidth, referring to the control bandwidth of the actuation systems. This typically refers to
actuator force generation bandwidth for all possible motions across the actuator mounting
points. The narrow bandwidth systems typically exhibit actuator control bandwidth around or
slightly above sprung mass frequency and would use the passive spring, positioned in series to
the actuator, to address high-frequency disturbances and vibrations. As such, their bandwidth
is typically between 3 and 7 Hz.[119] Broad bandwidth systems, on the other hand, attempt
to address oscillations beyond the sprung mass frequency using high-bandwidth actuators
typically above 10 Hz.[119]
Since the broad bandwidth active suspensions (BBASs) attempt to control both the slower
body modes and the fast wheel-hop modes, it is generally expected that the broad bandwidth
systems will be more challenging to implement. It is expected that such suspensions will
incur higher costs, complexities, and possible fuel consumption.[120] However, one should
1053
also observe that narrow bandwidth systems may experience similar fuel consumption due
to either their higher compliance and the corresponding increased pump flow/displacement
requirement [119] or higher control effort associated with the mismatch between the desired
and actual control bandwidth.[43]
5.5.1.
Narrow bandwidth active suspensions
Some narrow bandwidth systems have been brought to the market with limited success. An
example is the Nissan Infinity Q45a [121] whose hardware included: (1) an accumulator (for
bandwidth limitation and NVH containment); (2) a mechanical supporting spring; and (3)
a relatively inexpensive pressure control valve. It had a single-acting actuator configuration
which is simple and inexpensive. It does, however, limit the system controllability on the
rebound stroke, which is in good part determined by the spring/accumulator settings.
Production hydraulic active suspension systems typically use pressure control valves to
regulate the desired force. Considering that a force-generating actuator is accompanied by
corresponding displacements and velocities of the in-series compliance in the total system,
one may consider the flow control valves to regulate the desired displacements/velocities.
Benefits and limitations of both force and displacement control approaches have been
discussed in [119,122124].
DaimlerChrysler introduced the Active Body Control (ABC) system in the 2000 CL
sedan. It consists of four low-bandwidth (5 Hz) hydraulic actuators providing active load
levelling, heave, pitch, and roll control. The ABC suspension system consists of [125,126]
hydraulic cylinders in series with steel coil springs (Figure 11). This cylinder sets the spring
preload and the series arrangement leads to the low-bandwidth system behaviour. In parallel with the spring/cylinder is a twin-pipe gas shock absorber (damper). Hydraulic fluid is
supplied to each corner via a pump regulated to 200 bars, and an accumulator at each axle
provides pressure levelling during peak loads. High-pressure fluid is carried by two valve
blocks one at each axle. Each block contains two 3-way proportional valves one for each
corner which are used to maintain plunger pressure and a 2-way shut-off valve for each corner to lock the cylinder when the vehicle is stopped and to provide failsafe operation. Return
fluid is damped through a return accumulator.
Figure 11. Schematic diagram of the Mercedes Active Body Control actuator.
1054
The ABC system utilises multiple sensors for mechanical state measurements. Sensors
include longitudinal and lateral accelerometers, three vertical accelerometers (for bump/
pitch/roll detections), as well as a level sensor and a plunger position sensor at each wheel.
Additionally, several hydraulic pressure sensors are used for pressure control and monitoring.
The ABC system employs various algorithm modules for vehicle dynamics control
depending on the vehicle states. Functions provided include speed-based load levelling and
ride height adjustment, skyhook control of the vehicle body heave mode, roll control for roll
angle response to vehicle lateral acceleration, and pitch control for pitch angle response to
longitudinal acceleration. The algorithms are further controlled by driver input via instrument panel height and performance mode switches. ABC has continued to evolve since its
first introduction in 1998. For example, the wheel damping is now continuously adjustable
and the efficiency of the pump has been increased. In addition, computational processing
power is more than double that of the previous system.[127]
5.5.2.
BBAS concept implementation at Ford
A broad bandwidth system was developed at Ford Research Laboratory in the early
1990s and was demonstrated by fitting the suspension concept hardware onto a 1989 Ford
Thunderbird.[128] The concept hardware and software verified the previous theoretical study
in ride quality improvement for rough road and for going over a speed bump or crossing
a crowned road. The effort also identified the shortcomings of the implemented hardware
structure to facilitate next generation active suspension hardware design.
This concept car development provided first-hand experience and useful answers to many
practical concerns such as actual power consumption, secondary ride harshness in the range
above 5 Hz, and actuator noise. It also enabled the identification of specific next steps to
address these issues. For example, some of the key items that have been identified to be
addressed in the future include the parasitic losses with the parallel spring structure and the
need for further refinement in the secondary ride quality.
The Ford Thunderbird BBAS research vehicle (Figure 12) used four high-fidelity electrohydraulic servo actuators, one at each corner of the vehicle. While a typical narrow bandwidth
suspension is controlled through three-way proportional pressure control valves, the BBAS
Figure 12. Ford BBAS concept implementation.
1055
was controlled through four-way servo valves, which have much higher precision and speed
of response. In addition, the BBAS actuators were based on double-acting cylinders capable
of equally fast rebound and jounce strokes. The vehicle also had one central and four cornerunit microprocessors for fast signal/control processing; four actuator displacement sensors
and four load cells for internal (force) loop calculations; and four air springs one at each
corner placed in parallel with the BBAS actuators. The air springs support and self-centre
the vehicle sprung mass as a typical load levelling set would do and at the same time provide
lower sprung mass natural frequency for more comfortable basic ride, which is then appropriately dynamically modified through the BBAS actuators. The system incorporated 26 various
sensors, including accelerometers, pressure sensors, vehicle speed sensors, and others.
The BBAS control strategy was based on coordinated individual wheel control and consisted of two hierarchical levels.[43,128] The outer loop level operated at a 20 ms rate. It
calculated the desired corner forces for the four BBAS actuators, desired operating modes
(handling or ride dominated), and checked the overall system integrity. The ride-related calculations were based on quarter-car vehicle models aimed at emulating skyhook damping
at each corner, which is often very close to the optimal possible ride benefit.[20] Different
effective spring and damping rates were used depending on prevailing operating modes, that
is, ride or handling. Table 1 summarises typical parameters used.[128]
The calculations also included roll and pitch attitude control during turning, braking, accelerating, and the like; this was based on lateral and longitudinal acceleration signals and
steering sensors. The commanded or desired levels of corner forces were communicated
to corner computers, which operated at higher sampling rates of 1 ms to implement the
inner-loop force control based on feedback provided by load cell sensors. The corner units
distributed the force requests between different servo valves. For example, the conventional
(relative) suspension damping and the passive segments of skyhook force were provided by
bypass or cross-port two-way servo valves that provided damping only when called for in a
manner similar to semi-active damping control.
The BBAS R&D vehicle accomplished many of its key objectives, especially with respect
to primary ride (up to circa 5 Hz) and handling. The typical results in terms of NASA ride
metrics are summarised in [128] where a smaller number on the scale of 10 implies better
ride. From this reference it can be seen that, on average, the active Thunderbird provided
the best ride among a dozen different vehicles under consideration. The only areas where the
BBAS had somewhat worse performance were on smooth roads, where Jaguar XJ6 and Lexus
LS 400 (both well known for their smooth ride) had an advantage. This may be attributable
to still less refined NVH (higher frequency) levels and related secondary ride quality that are
some of the remaining challenges for BBAS concepts. Based on their summarised results,
it should also be noted that the BBAS Thunderbird significantly outperformed both the
baseline passive suspension Thunderbird and its narrow bandwidth (NBAS) counterparts as
exemplified by production Infiniti Q45a and Toyota Soarer.
Table 1. Typical parameters used in Ford BBAS.
Ride mode
Parameter
Skyhook damping
Units
Lb-s/in
N-s/m
Conventional damping Lb-s/in
N-s/m
Suspension spring
Lb/in
N/m
Manoeuvring mode
Front
Rear
Front
Rear
23.5
4115.3
3.0
525.4
15.0
2626.8
10.0
1751.2
3.0
525.4
15.0
2626.8
23.5
4115.3
8.0
1400.9
120.0
21,014
10.0
1751.2
4.0
700.5
120.0
21,014
1056
In the same paper, the handling characteristics of the active Thunderbird are compared with
its passive (production) version. The results of back-to-back tests show that the active vehicle
results in more agile handling with faster yaw response, less roll and pitch, and significantly
larger maximum lateral acceleration (0.895 vs. 0.81 g). Thus, in a direct comparison between
the two vehicles, the active car displayed significant improvements in both vehicle ride and
handling.
While achieving the full potential of a BBAS still remains an elusive goal, significant
progress has been achieved with the above early efforts. This holds particularly true for the
case of primary ride and handling. The challenging areas in need of further development
include secondary ride, NVH, and system power consumption, complexity, and cost. Possible
approaches here have been outlined in many of the previously mentioned references (e.g.
[7,11,88,107,119]).
In closing, it should also be mentioned that in addition to the above electro-hydraulic
implementation of the BBAS there was an additional BBAS effort focused on Electrical
Active Suspension (EAS), which was capable of power/energy regeneration.[129] The EAS
concept was also successfully implemented in a prototype vehicle with similar results and
challenges as for the electro-hydraulic counterpart.
5.6.
Road preview feature using road information for semi-active and active control
The concept of utilising the road profile information prior to road disturbances hitting the
wheel, investigated in [3135], has been partially applied to production vehicles. The early
attempt includes the Nissan vehicles in 1990 with the Super Sonic Suspension [130] feature.
More recently, the preview feature can be found in the 2013 Mercedes S-Class with the
Magic Body Control feature.[131,132] The early Super Sonic Suspension was applied to a
semi-active suspension system and the recent Magic Body Control is equipped with an active
suspension.
The early Nissan Super Sonic Suspension had only three levels of damping that are automatically selected to provide comfort and handling based on the road surface profile scanned
with ultrasonic sensors, as well as steering angle, vehicle speed, brake on/off status, and
measured accelerations.
The Mercedes S-Class detects road surface undulations in advance using a stereo camera
and adjusts the suspension to deal with them accordingly. The suspension system adjusts its
shock absorber damping in advance, making it stiffer or softer for each individual wheel, and
uses active hydraulic pistons to control the desired wheel loads.
The preview information for a three-dimensional road profile is obtained through a stereo
camera that monitors an area up to 15 m ahead during good visibility preferably in the day
light with suitable road surface structures and at speeds up to 130 kph. The stereo camera
has an update rate of about 60 ms. Each section of the road can be measured several times
at various angles during vehicle approach, and statistical algorithms are used to provide the
final estimate of road height profile. It has been demonstrated that, under good visibility, the
system performs well, providing superior ride comfort [133] especially when going through
speed bumps and excellent handling during dynamic manoeuvres.
6.
Conclusions
A 2 DOF quarter-car model, equipped with an actuator that generates force between the
car body/sprung mass and wheel/unsprung mass, has been widely studied in the literature
1057
and in the process facilitated many useful insights and new concepts (sky hook, ground
hook, invariant points to name a few). Thus, this simple model served well as a base for
the comparison study of various suspension concepts, controller designs, benefits, and limitations. The present survey first discusses the various performance index/cost functions that
are important for the design of a passenger vehicle. It then reviews controller design limitations through the application of optimal control theory to a quarter-car model with an
LQ-based cost function. Figures are generated to illustrate the design limitations and tradeoffs in RMS measures of various vehicle attributes and their associated frequency contents.
These plots should help suspension designers to visualise the ability of active suspensions to
behave differently depending on the current (and forthcoming) road excitations and vehicle
manoeuvre-induced loading (e.g. braking, accelerating, and turning).
Based on more than 40 years of theoretical and practical developments in the industry and
academia, it can be concluded that one of the main challenges for widespread usage of active
suspensions still lies in the area of actuator design and implementation. It is expected that
superior outcome will eventually result from a synergetic interplay between control software
intelligence and ingenious hardware design, all led by model-based system engineering and
further facilitated by relentless developments in electronics, microcontrollers and associated
sensors, and vehicle-to-vehicle and vehicle-to-infrastructure connectivity. Indeed, an important part of such an actuator development should include realisation of the desired actuator
force or, alternatively, displacement. As mentioned before, producing and tracking the desired
force (or displacement) is by no means a trivial task, especially when combined with practical constraints on cost, packaging, weight, and energy consumption. This important aspect
of advanced suspension design should be carefully evaluated through computer simulations,
bench testing, and eventually full vehicle on-road verification.
The above review of current state of the art reveals that the next advances in active
and semi-active suspension design will mainly come from two thrust areas. The first is the
increased efficiency in actuator design and implementation such as the usage of systematic
control software algorithm design combined with the previously mentioned innovative hardware design measures such as inclusion of possibly variable in-series/parallel compliances
and fast load levelling. The second is more comprehensive usage of preview information from
camera, Global Positioning System (GPS), and electronic horizon such as vehicle-to-vehicle
communication, vehicle-to-infrastructure communication, vehicle localisation and real-time
accessing of cloud information, and crowd sourcing leading to up-to-date road profile maps.
Acknowledgement
The authors would like to thank their colleagues Li Xu, Vladimir Ivanovic, and Michael Fodor for their valuable
input and feedback on this survey.
Disclosure statement
No potential conflict of interest was reported by the authors.
References
[1] Galip UA, Peng H, Melih C. Automotive control systems. New York (NY): Cambridge University Press;
2012.
[2] Cao D, Song X, Ahmadian M. Editors perspectives: road vehicle suspension design, dynamics, and control.
Veh Syst Dyn. 2011;49(12):328.
[3] Lu J, Hrovat D, Thomas P, Engleman J, Tseng HE, Filev D. Adaptive crash height adjustment using active
suspensions. US patent 13/736,397. 2013.
1058
[4] Tseng EH, Hrovat D, Lu J, McConnell M, Dehmel M, Seemann M, Owens FV. Suspension control system to
facilitate wheel motions during parking. US Patent 8,825,292. 2014 Sep 2.
[5] Smith MC, Walker GW. Performance limitations and constraints for active and passive suspension: a
mechanical multi-port approach. Veh Syst Dyn. 2000;33(3):137168.
[6] Hrovat D. A class of active LQG optimal actuators. Automatica. 1982;18(1):117119.
[7] Hrovat D. Survey of advanced suspension developments and related optimal control applications. Automatica.
1997;33(10):17811817.
[8] Karnopp D, Crosby M, Harwood R. Vibration control using semi-active force generators. J Manuf Sci Eng.
1974;96(2):619626.
[9] Wilson D, Sharp R, Hassan S. The application of linear optimal control theory to the design of active
automobile suspensions. Veh Syst Dyn. 1986;15(2):105118.
[10] Yue C, Butsuen T, Hedrick JK. Alternative control laws for automotive active suspensions. In American
Control Conference; 1988 June 1517; Atlanta, GA.
[11] Sharp RS, Peng H. Vehicle dynamics applications of optimal control theory. Veh Syst Dyn. 2011;49(7):1073
1111.
[12] Smith CC, McGehee DY, Healey AJ. The prediction of passenger riding comfort from acceleration data.
ASME J Dyn Syst Meas Control. 1978;100(1):3441.
[13] Hedrick JK, Ravera RK, Amderes JR. The effect of elevated guideway construction tolerances on vehicle ride
quality. ASME J Dyn Syst Meas Control. 1975;97(4):408416.
[14] Fearnsides JJ, Hedrick JK, Firouztash H. Specification of ride quality criteria for transportation systems: the
state of the art and a new approach. High Speed Ground Transp J. 1974;8(2):125132.
[15] Hoberock LL. A survey of longitudinal acceleration comfort studies in ground transportation vehicles. ASME
J Dyn Syst Meas Control. 1977;99(2):7684.
[16] Anonymous. ISO: Mechanical vibration and shock: evaluation of human exposure to whole-body vibration.
ISO 2631, International Standard Organization; 1997.
[17] Smith CC. On using the ISO standard to evaluate the ride quality of broad-band vibration spectra in
transportation vehicles. ASME J Dyn Syst Meas Control. 1976;98(4):440443.
[18] Stephens GD. Comparative vibration environments of transportation vehicles. In: Proceedings of the Design
Engineering Technical Conference. Passenger Vibration in Transportation Vehicles; 1977 Sept 2628;
Chicago, IL. p. 5972.
[19] Hrovat D. Applications of optimal control to advanced automotive suspension design. ASME J Dyn Syst
Meas Control. 1993;115(2B):328342.
[20] Hrovat D. Chapter 31: Active and semi-active suspension control. In: Mastinu G, Plchl M, editors. Road and
off-road vehicle system dynamics handbook. Boca Raton (FL): CRC Press, Taylor and Francis Group; 2014.
p. 11791226.
[21] Boyd S, El Ghaoui L, Feron E, Balakishnan V. Linear matrix inequalities in system and control theory.
Philadelphia (PA): SIAM; 1994.
[22] Fialho I, Balas GJ. Road adaptive active suspension design using linear parameter-varying gain-scheduling.
IEEE Trans Control Syst Technol. 2002;10(1):4354.
[23] Becker G, Packard A. Robust performance of linear parametrically varying systems using parametricallydependent linear feedback. Syst Control Lett. 1994;23(3):205215.
[24] Chen H, Guo K. An LMI approach to multiobjective RMS gain control for active suspensions. In: Proceedings
of the 2001 American Control Conference; 2001 June 2527; Arlington, VA. p. 26462651.
[25] Tran M, Hrovat D. Application of gain-scheduling to design of active suspensions. In: Proceedings of the
1993 IEEE Conference on Decision and Control. Vol. 2; 1993 Dec 1517; San Antonio, TX. p. 10301035.
[26] Hedrick JK, Butsuen T. Invariant properties of automotive suspensions. Proc Inst Mech Eng D: J Automob
Eng. 1990;204(1):2127.
[27] Hrovat D, Margolis D. Realistic road-track systems simulation using digital computers. In: Proceedings of the
Winter Simulation Conference; 1975 Dec; Sacramento, CA. p. 481489.
[28] Hrovat D. Optimal passive vehicle suspensions [Ph.D. thesis]. Davis (CA): University of California;
1979.
[29] Tseng HE, Hedrick JK. Semi-active control laws-optimal and sub-optimal. Veh Syst Dyn. 1994;23(1):545
569.
[30] Giorgetti N, Bemporad A, Tseng HE, Hrovat D. Hybrid model predictive control application towards optimal
semi-active suspension. Int J Control. 2006;79(5):521533.
[31] Tomizuka M. Optimum linear preview control with application to vehicle suspension revisited. ASME J
Dyn Syst Meas Control. 1976;98:309315.
[32] Louam N, Wilson DA, Sharp RS. Optimal control of a vehicle suspension incorporating the time delay
between front and rear wheel inputs. Veh Syst Dyn. 1988;17(6):317336.
[33] Louam N, Wilson DA, Sharp RS. Optimization and performance enhancement of active suspensions for
automobiles under preview of the road. Veh Syst Dyn. 1992;21(1):3963.
[34] Hrovat D. Optimal suspension performance for 2D vehicle models. J Sound Vib. 1991;146(1):
93110.
[35] Hac A. Optimal linear preview control of active vehicle suspension. Veh Syst Dyn. 1992;21:167195.
[36] Hac A, Youn I. Optimal semi-active suspension with preview based on a quarter car model. J Vib Acoust.
1992;114(1):8492.
1059
[37] Hac A, Youn I. Optimal design of active and semi-active suspensions including time delays and preview. J
Vib Acoust. 1993;115(4):498508.
[38] Van Auken RM. Active suspension control synthesis using reduced order models that account for the time
delay between road inputs. In: ASME 8th Biennial Conference on Engineering Systems Design and Analysis;
2006 July 47; Torino, Italy. p. 451462.
[39] Jankovic M, Kolmanovsky I. Developments in control of time-delay systems for automotive powertrain
applications. In: Balachandran B, Kalmr-Nagy T, Gilsinn DE, editors. Delay Differential Equations. Berlin:
Springer; 2009. p. 5592.
[40] Thompson AG, Pearce C, Millers E. Direct computation of the performance index for an optimally controlled
active suspension with preview applied to a half-car model. Veh Syst Dyn. 2001;35(2):121137.
[41] Pilbeam C, Sharp RS. On the preview control of limited bandwidth vehicle suspensions. Proc Inst Mech Eng
D: J Automob Eng. 1993;207(3):185193.
[42] Prokop G, Sharp RS. Performance enhancement of limited-bandwidth active automotive suspensions by road
preview. IEE Proc Control Theory Appl. 1995;142(2):140148.
[43] Goran MB, Bachrach BI, Smith RE. The design and development of a broad bandwidth active suspension
concept car. In: XXIV FISITA Congress. Total Vehicle Dynamics. Vol. 2, no. IMECHE C389/418; 1992.
p. 231252.
[44] Smith MC. Achievable dynamic response for automotive active suspension. Veh Syst Dyn. 1995;24(1):
133.
[45] Smith MC, Wang FC. Controller parameterization for disturbance response decoupling: application to vehicle
active suspension control. IEEE Trans Control Syst Technol. 2002;10(3):393407.
[46] Wang FC, Smith MC. Disturbance response decoupling and achievable performance with application to
vehicle active suspension. Int J Control. 2002;75(12):946953.
[47] Brezas P, Smith MC. Linear quadratic optimal and risk-sensitive control for vehicle active suspensions. IEEE
Trans Control Syst Technol. 2014;22(2):543556.
[48] Hrovat D, Hubbard M. Optimum vehicle suspensions minimizing RMS rattlespace, sprung-mass acceleration
and jerk. J Dyn Syst Meas Control. 1981;103(3):228236.
[49] Davis BR, Thompson AG. Optimal linear active suspensions with integral constraint. Veh Syst Dyn.
1988;17(6):357366.
[50] Youn I, Im J, Tomizuka M. Level and attitude control of the active suspension system with integral and
derivative action. Veh Syst Dyn. 2006;44(9):659674.
[51] Karnopp D. Force generation in semi-active suspensions using modulated dissipative elements. Veh Syst Dyn.
1987;16(56):333343.
[52] Lin J, Kanellakopoulos I. Nonlinear design of active suspensions. IEEE Control Syst. 1997;17(3):
4559.
[53] Chantranuwathana S, Peng H. Adaptive robust force control for vehicle active suspensions. Int J Adapt
Control Signal Process. 2004;18(2):83102.
[54] Gordon TJ, Marsh C, Wu QH. A learning automaton methodology for control system design in active
vehicle suspensions. In: International Conference on Control 94. Vol. 1; 1994 Mar 2124; Coventry, UK.
p. 326331.
[55] Howell MN, Frost GP, Gordon TJ, Wu QH. Continuous action reinforcement learning applied to vehicle
suspension control. Mechatronics. 1997;7(3):263276.
[56] Valasek M, Kortum W. Nonlinear control of semi-active road-friendly truck suspension. In: Proceedings of
AVEC. Vol. 98; 1998 Sep 1418; Nagoya, Japan. p. 275280.
[57] Gordon TJ, Best MC. Dynamic optimization of nonlinear semi-active suspension controllers. In: International
Conference on Control 94. Vol. 1; 1994 March 2124, Coventry, UK. p. 332337.
[58] Gordon TJ, Sharp RS. On improving the performance of automotive semi-active suspension systems through
road preview. J Sound Vib. 1998;217(1):163182.
[59] Wang J, Wilson DA. Mixed GL2/H2/GH2 control with pole placement and its application to vehicle
suspension system. Int J Control. 2001;74(13):13531369.
[60] Abdellahi E, Mehdi D, MSaad M. On the design of active suspension system by Hinf and mixed H2/Hinf:
an LMI approach. In: Proceedings of the 2000 American Control Conference; 2000 June 2830; Chicago, IL.
p. 40414045.
[61] Lu J, DePoyster J. Multi-objective optimal suspension control to achieve integrated ride and handling
performance. IEEE Trans Control Syst Technol. 2002;10(6):807821.
[62] De Jager AG. Comparison of two methods for the design of active suspension systems. Optim Control Appl
Methods. 1991;12(3):173188.
[63] Yamashita M, Fujimoti K, Uhlik C, Kawatani R, Kumura H. Hinf control of an automotive active suspension.
In: Proceedings of the 29th IEEE Conference on Decision and Control; 1990 Dec 57; Honolulu, HI. p. 2244
2250.
[64] Balas GJ, Doyle JC, Glover K, Packard A, Smith R. Mu-analysis and synthesis toolbox, Matlab users guide.
Boston (MA): MathWorks Inc; 1993.
[65] Doyle J, Packard A, Zhou K. Review of LFTs, LMIs, and Mu. In: Proceedings of the 30th IEEE Conference
on Decision and Control. Vol. 2; 1991 Dec 1113; Brighton, UK. p. 12271232.
[66] Gobbi M, Mastinu G. Analytical description and optimization of the dynamic behaviour of passively
suspended road vehicles. J Sound Vib. 2001;245(3):457481.
1060
[67] Chatillon M, Jezequel L, Coutant P, Baggio P. Hierarchical optimization of the design parameters of a vehicle
suspension system. Veh Syst Dyn. 2006;44(11):817839.
[68] Chen H, Sun P, Guo K. A multi-objective control design for active suspensions with hard constraints. Proc
Am Control Conf. 2003;5:43714376.
[69] Georgiou G, Verros G, Natsiavas S. Multi-objective optimization of quarter-car models with a passive or
semi-active suspension system 45. Veh Syst Dyn. 2007;45(1):7792.
[70] Crews J, Mattson M, Buckner G. Multi-objective control optimization for semi-active vehicle suspensions. J
Sound Vib. 2011;330(23):55025516.
[71] Yi K, Hedrick J. Active and semi-active heavy truck suspensions to reduce pavement damage. SAE Technical
Paper 892486; 1989.
[72] Cole DJ, Cebon D. Truck suspension design to minimize road damage. Proc Inst Mech Eng D: J Automob
Eng. 1996;210(2):95107.
[73] Kitching KJ, Cole DJ, Cebon D. Theoretical investigation into the use of controllable suspensions to minimize
road damage. Proc Inst Mech Eng D: J Automob Eng. 2000;214(1):1331.
[74] Cole DJ. Fundamental issues in suspension design for heavy road vehicles. Veh Syst Dyn. 2001;35(45):
319360.
[75] Novak M, Valasek M. A new concept of semi-active control of trucks suspension. In: Proceedings of
AVEC96; 1996 June 2428; Aachen, Germany. p. 141152.
[76] Valek M, Novak M, ika Z, Vaculin O. Extended ground-hook-new concept of semi-active control of trucks
suspension. Veh Syst Dyn. 1997;27(56):289303.
[77] Valek M, Kortm W, ika Z, Magdolen L, Vaculn O. Development of semi-active road-friendly truck
suspensions. Control Eng Pract. 1998;6(6):735744.
[78] Vaculn O, Valek M, Kortm W. Multi-objective control of semi-active truck suspensions. In: Proceedings
of AVEC2000; 2000 August 2224; Ann Arbor, MI. p. 405412.
[79] Valasek M, Sveda J, Sika Z. Soil-friendly off-road suspension. Veh Syst Dyn. 1998;44(Suppl 1):
479488.
[80] Valasek M, Kejval J, Maca J, Smilauer V. Bridge-friendly truck suspension. In: Proceedings of the 18th
IAVSD Symposium. Vol. 41; 2003 Aug 2529; Atsugi, Japan. p. 1322.
[81] MVO, Valek KJ. Global chassis control: integration synergy of brake and suspension control for active
safety. In: Proceedings of the International Symposium on Advanced Vehicle Control; 2004 Aug 2327;
Arnhem, The Netherlands. p. 495500.
[82] van Zanten A. Control of horizontal vehicle motion. In: Mastinu G, Plchl M, editors. Road and offroad vehicle system dynamics handbook. Boca Raton (FL): CRC Press, Taylor and Francis Group; 2014.
p. 10931178.
[83] Papageorgiou C, Smith MC. Positive real synthesis using matrix inequalities for mechanical networks:application to vehicle suspension. IEEE Trans Control Syst Technol. 2006;14(3):423435.
[84] Smith MC. Synthesis of mechanical networks: the inerter. IEEE Trans Automat Control. 2002;47(10):
16481662.
[85] Karnopp DC, Margolis DL, Rosenberg RC. System dynamics: a unified approach. 3rd/8th ed. New York:
Wiley; 2000 (3rd)/ 2012 (8th).
[86] Margolis D, Wilson R. Bond graph modeling of fluid filled engine mounts. Simulation Series 31: Proceedings
of the 1999 International Conference on Bond Graph Modeling and Simulation (ICBGM 99): [held as part
of] 1999 Western Multi Conference; 1999 January 1720; San Francisco, CA. p. 211216.
[87] Singh R, Adiguna H, Tiwari M, Tseng H, Hrovat D. Transient response evaluation of a hydraulic engine
mount. J Sound Vibration. 2002;36(7):3035.
[88] Evangelou S, Limebeer DJN, Sharp RS, Smith MC. Control of motorcycle steering instabilities. IEEE Control
Syst Mag. 2006;26(5):7888.
[89] Smith MC, Wang FC. Performance benefits in passive vehicle suspensions employing inerters. Veh Syst Dyn.
2004;42(4):235257.
[90] Gillespie TD. Fundamentals of vehicle dynamics. Warrendale (PA): Society of Automotive Engineers;
1992.
[91] Malek KM, Hedrick JK. Decoupled active suspension design for improved automotive ride quality/handling
performance. Veh Syst Dyn. 1986;15(Suppl 1):383398.
[92] Hayakawa K, Matsumoto K, Yamashita M, Suzuki Y, Fujimori K, Kimura H. Robust H-output feedback control of decoupled automobile active suspension systems. IEEE Trans Automat Control. 1999;44(2):
392396.
[93] Venhovens PJT. Optimal control of vehicle suspensions [Ph.D. Thesis]. Delft: Delft University of Technology;
1993.
[94] Best M, Gordon T, Dixon P. An extended adaptive Kalman filter for real-time state estimation of vehicle
handling. Veh Syst Dyn. 2000;34:5775.
[95] Jeppesen BP, Cebon D. Analytical redundancy techniques for fault detection in an active heavy vehicle
suspension. Veh Syst Dyn. 2004;42(12):7588.
[96] Nastasic , Dek Jahn G. Citroen technical guide [Internet]. Available from: http://www.citroenkerho.fi/xantia
/pdf/tekniikka/Tekniikkaopas.pdf
[97] Chance BK. 1984 Continental mark VII/Lincoln continental electronically-controlled air suspension (EAS)
System. SAE Technical Paper 840342; 1984.
1061
[98] Edmunds D. 2012 Tesla Model S signature performance suspension walkaround [Internet]. 2012 Sep 26 [cited
2015 Mar]. Available from: http://www.edmunds.com/car-reviews/track-tests/2012-tesla-model-s-signatureperformance-suspension-walkaround.html
[99] Korosec K. Potholes and Teslas model S: never the twain shall meet [Internet]. 2014 Sep 19 [cited 2015
Mar]. Available from: http://fortune.com/2014/09/19/tesla-model-s-suspension-upgrade/
[100] Kojima H, Nakano J, Nakayama H, Kawashima N, Fujimoto H. Development of new Toyota electronic
modulated suspension-two concepts for semi-active suspension control. SAE Technical Paper 911900;
1991.
[101] Nicolas R. Intelligent suspension system of Lincoln MKZ help mitigating pothole damage [Internet].
2014 Mar 20. Available from: http://www.car-engineer.com/intelligent-suspension-system-lincoln-mkz-helpmitigating-pothole-damage/
[102] Lincoln Drive Control: CCD. Lincoln Motor Company [Internet]. Available from: http://www.lincoln.com/
cars/mks/features/?featID = 1#page = Feature1
[103] Bodie MO, Hac A. Closed loop yaw control of vehicles using magneto-rheological dampers. SAE Technical
Paper 2000-01-0107; 2000.
[104] Hac A, Bodie MO. Improvements in vehicle handling through integrated control of chassis systems. Int J Veh
Autonomous Syst. 2002;1(1):83110.
[105] Fodor M, Redfield RC. Resistance control, semi-active damping performance. In: ASME Dynamic Systems
and Control Division, International Mechanical Engineering Congress and Exposition; 1995 Nov. 1217; San
Francisco, CA. p. 161169.
[106] Fodor M, Redfield RC. Experimental verification of resistance control, semi-active damping. Veh Syst Dyn.
1996;26(2):143159.
[107] Sharp RS. Variable geometry active suspension for cars. Comput Control Eng J. 1998;9(5):217222.
[108] Watanabe Y, Sharp RS. Mechanical and control design of a variable geometry active suspension system. Veh
Syst Dyn. 1999;32(23):217235.
[109] Tumova G. Delft active suspension (DAS-II) [Internet]. Research report OND1281145. Available from:
http://www.narcis.nl/research/RecordID/OND1281145
[110] Venhovens PT, van der Knaap AM. Delft active suspension (DAS) background theory and physical realization. In: Pauwelussen JP, Pacejka HB, editors. Smart vehicles. Amsterdam: Swets & Zeitlinger Publishers;
1995. p. 139165.
[111] van der Knapp A. Design of a low power anti-roll/pitch system for a passenger car. Delft University of
Technology, Vehicle Research Laboratory, Report 89.3VT.2628.
[112] Venhovens P, van der Knaap A, Pacejka H. Semiactive vibration and attitude control. In: Proceedings of the
International Symposium on Advanced Vehicle Control, AVEC 92; 1992 Sept 1417; Yokohama, Japan.
p. 170175.
[113] Evers WJ, Besselink IJM, van der Knaap ACM, Nijmeijer H. Analysis of a variable geometry active suspension. In: Proceeding of International Symposium on Advanced Vehicle Control; 2008 Oct 69; Kobe
International Conference Center, Kobe, Japan. p. 350355.
[114] Packer MB. Active ride control a logical step from static vehicle attitude control. SAE Technical Paper
780050; 1978.
[115] Crolla DA, Horton DNL, Pitcher RH, Lines JA. Active suspension control for an off-road vehicle. Proc Inst
Mech Eng D: J Automob Eng. 1987;201(1):110.
[116] Raad J, Oliver J, Norton R, Kokotovic V. Method of improving response time in a vehicle active tilt control
system. US Patent 5,948,028. 1999.
[117] Raad J, Oliver J, Rhode D, Norton R, Hermann S, Watson J. Method of controlling a vehicle active tilt control
system. US Patent 6,076,027. 2000.
[118] Konik D, Bartz R, Barnthol F, Bruns H, Wimmer M. Dynamic drive the new active roll stabilization system from BMW group system description and functional improvements. In: Proceeding of International
Symposium on Advanced Vehicle Control; 2000 Aug 2224; Ann Arbor, MI. p. 413420.
[119] Hrovat D, Tseng E, Fodor M, Asgari J. Chapter 22: Active and semiactive suspension systems. In: Mastinu
G, Plchl M, editors. Road and off-road vehicle system dynamics handbook. Boca Raton (FL): CRC Press,
Taylor and Francis Group; 2014. p. 769796.
[120] Hillebrecht P, Konik D, Pfeil D, Wallentowitz H, Zieglmeier F. The active suspension between customer
benefit and technological competition. In: Proceedings of XXlV FISITA Congress. Total Vehicle Dynamics.
Vol. 2; 1992 June 711; London. p. 221230.
[121] Akatsu Y, Fukushima N, Talcahashj K, Satch M, Kawaranki Y. An active suspension employing an
electrohydraulic pressure control system. SAE Technical Paper 905123; 1990.
[122] Alleyne A, Liu R. On the limitations of force tracking control of hydraulic servosystems. ASME J Dyn Syst
Meas Control. 1999;121(2):184190.
[123] Zhang Y, Alleyne A. A practical and effective approach to active suspension control. In: Proceedings
of the 6th International Symposium on Advanced Vehicle Control; 2002 Sep 913; Hiroshima, Japan.
p. 153158.
[124] Peng H, Strathearn R, Ulsoy A. A novel active suspension design technique simulation and experimental
results. In: Proceedings of the 1997 American Control Conference. Vol. 1; 1997 June 46; Albuquerque, NM.
p. 709713.
1062
[125] Merker T, Gaston G, Olaf T. Active Body Control (ABC) the DaimlerChrysler active suspension and damping
system. SAE Technical Paper 2002-21-0054; 2002.
[126] Merker T, Wirtz J, Hill M, Jeglitzka M. Das SL Fahrwerk-Dynamik und Komfort vereint (The SL chassisdynamics and comfort combined). ATZ/MTZ Magazine Special Issue; 2001. p. 8491.
[127] Kruse J. Mercedes-Benz NEXT online magazine [Internet]. 2014 June 4. Available from: http://next.mercedes
-benz.com/en/mbc-eng/
[128] Goran M, Smith R. Insights gained from active suspension development. In: Proceedings of the 26th FISITA
Congress; 1996 June 1623; Prague, Czech Republic. p. 24862514.
[129] Davis RI, Patil PB. Electrically powered active suspension for a vehicle. US Patent 5,060,959; 1991 Oct 29.
[130] Sugasawa F, Kobayashi H, Kakimoto T, Shiraishi Y, Tateishi Y. Electronically controlled shock absorber
system used as a road sensor which utilizes supersonic waves. SAE Technical Paper 851652; 1985.
[131] Streiter IR. Active preview suspension system. ATZ Worldwide. 2008;110(5):411.
[132] Rauh J, Ammon D. System dynamics of electrified vehicles: some facts, thoughts, and challenges. Veh Syst
Dyn. 2011;49(7):10051020.
[133] Mercedes-Benz S550. Consumer reports; 2014 Sep. p. 53.
This article was downloaded by: [Syracuse University Library]

On: 04 June 2014, At: 16:55
Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered
office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK
Vehicle System Dynamics: International

Journal of Vehicle Mechanics and
Mobility
Publication details, including instructions for authors and
subscription information:
http://www.tandfonline.com/loi/nvsd20
Validation of vehicle dynamics

simulation models a review
a
Emir Kutluay & Hermann Winner
Department of Mechanical Engineering, Hacettepe University,

06800 Beytepe, Ankara, Turkey
b
FZD, TU Darmstadt, Petersenstrae 30, 64287, Darmstadt,

Germany
Published online: 13 Jan 2014.
To cite this article: Emir Kutluay & Hermann Winner (2014) Validation of vehicle dynamics
simulation models a review, Vehicle System Dynamics: International Journal of Vehicle Mechanics
and Mobility, 52:2, 186-200, DOI: 10.1080/00423114.2013.868500
PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the
Content) contained in the publications on our platform. However, Taylor & Francis,
our agents, and our licensors make no representations or warranties whatsoever as to
the accuracy, completeness, or suitability for any purpose of the Content. Any opinions
and views expressed in this publication are the opinions and views of the authors,
and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content
should not be relied upon and should be independently verified with primary sources
of information. Taylor and Francis shall not be liable for any losses, actions, claims,
proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or
howsoever caused arising directly or indirectly in connection with, in relation to or arising
out of the use of the Content.
This article may be used for research, teaching, and private study purposes. Any
substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,
systematic supply, or distribution in any form to anyone is expressly forbidden. Terms &
Downloaded by [Syracuse University Library] at 16:55 04 June 2014
Conditions of access and use can be found at http://www.tandfonline.com/page/termsand-conditions
Vehicle System Dynamics, 2014

Validation of vehicle dynamics simulation models a review
Emir Kutluaya and Hermann Winnerb

a Department of Mechanical Engineering, Hacettepe University, 06800 Beytepe, Ankara, Turkey;
b FZD, TU Darmstadt, Petersenstrae 30, 64287, Darmstadt, Germany
(Received 12 September 2013; accepted 18 November 2013 )

In this work, a literature survey on the validation of vehicle dynamics simulation models is presented.
Estimating the dynamic responses of existing or proposed vehicles has a wide array of applications in
the development of vehicle technologies, e.g. active suspensions, controller design, driver assistance
systems, etc. Although simulation environments, measurement tools and mathematical theories on
vehicle dynamics are well established, the methodical link between the experimental test data and
validity analysis of the simulation model is still lacking. This report presents different views on the
definition of validation, and its usage in vehicle dynamics simulation models.
Keywords: validation; simulation; modelling; vehicle dynamics
1.
Introduction
The simulation of vehicle dynamics has a wide array of applications in the development of
vehicle technologies, i.e. active suspensions, chassis design, controller design, driver assistance systems, development of simulators for ergonomics research, etc. Vehicle dynamics
simulations reduce the duration and costs during the research and development stages of new
designs and technologies.
Although simulation environments, measurement tools and mathematical theories on vehicle dynamics are well established, the methodical link between the experimental test data
and validity analysis of the simulation model is still lacking. This thesis aims to introduce a
methodology to be used in assessment of vehicle dynamics simulation models.
Validation of vehicle dynamics simulations is an intersection of two fields of study: simulation of vehicle dynamics and validation of simulations. Thus, the literature domain can be
divided into three main subjects: verification and validation (V&V) of computational models;
validation of vehicle dynamics simulation models in practice and validation methodologies
for vehicle dynamics simulation models.
In this paper, first the literature on V&V of computational models is presented. Definitions of
the V&V concepts, and approaches to the question from different disciplines are explored. Next
vehicle dynamics and the utilisation of simulation models in vehicle dynamics are explored.
The final part of the literature survey deals with the validation studies dedicated to simulation
of vehicle dynamics.
Corresponding
author. Emails: kutluay@hacettepe.edu.tr; emir.kutluay@gmail.com
2014 Taylor & Francis
187
2. V&V of computational models
2.1.
Definitions of concepts of V&V
Many definitions on V&V can be found in the literature. The most important concept and
definitions by various authors are presented in this subsection.
The goal of the V&V is to find out if a model is accurate when used to predict the performance
of the real-world system that it represents, or to predict the difference in performance between
two scenarios or two or more model configurations.[1] The process of verifying and validating
a model should also lead to improving a models credibility with users [2] and decisionmakers.[3]
Verification of a simulation model, in laymans terms, is defined as building the model
right. According to Carson,[1] verification is the techniques that are used to assure that
the model is correct and matches any agreed upon specifications and assumptions. Another
similar definition is the process of determining if a computational model of a physical event
and the code implementing the computational model can be used to represent the mathematical
model of the event with sufficient accuracy.[3] In the field of computational fluid dynamics
(CFD), the American Institute of Aeronautics and Astronautics (AIAA) definition is generally
accepted: The process of determining that a model implementation accurately represents the
developers conceptual description of the model and the solution to the model.[4] In another
source concerning CFD applications,[5] verification is defined as a process for assessing
simulation numerical uncertainty and estimating the sign, magnitude and uncertainty of the
simulation numerical error.
Same richness of definitions can also be encountered for the validation concept. Similar
to verification, in laymans terms, validation means Building the right model. One of the
earliest definitions is the process of confirming that the conceptual model is applicable or
useful by demonstrating an adequate correspondence between the computational results and
the actual other theoretical data.[6] Validation can be defined as the processes and techniques
that the model developer, model customer and decision-makers jointly use to assure that the
model represents the real system (or proposed real system) to a sufficient level of accuracy.[1]
Validation is the substantiation that a computerised model within its domain of applicability
possesses a satisfactory range of accuracy consistent with the intended application of the
model.[7] It is the process of determining if a mathematical model of a physical event represents
the actual physical event with sufficient accuracy.[3] According to AIAA,[4] validation is the
process of determining the degree to which a model is an accurate representation of the real
world from the perspective of the intended uses of the model.
Absolute validity is refuted by many experts.[2,3,8,9] A models validity is only defined
within the limits of the project and the intended application. Although a more comprehensive
validity analysis increases the credibility of the model, it also comes with extra financial and
time cost. Thus, a simulation model of a complex system can only be an approximation of the
actual system.[8] The logical conclusion is that, no matter how much time is spent to develop,
enhance and validate the model, there will always be discrepancies between the physical
phenomenon to be modelled and the simulation results.
2.2.
Philosophical aspect
The main question of concern is Can a simulation model be validated? when the philosophical perspectives to the simulation and validation are considered. This question is actually
very similar to one of the main problems of philosophy of science, regarding the scientific
progression.
188
E. Kutluay and H. Winner
As previously stated, pure and absolute validation is impossible. According to Popper,[10]

scientific theories cannot be proven; they can only be tested through observations. Therefore,
a simulation model can only be invalidated when the performance of the model fails to meet
the accuracy criteria, but cannot be validated otherwise. If the simulation model fulfils the
defined validity criteria, then it can be deemed not invalid under the defined specific set of
operating conditions and limits.
Comparison of different perspectives on philosophical aspects of validation is examined by
Klein and Herskovitz.[11]
2.3. Approaches to V&V

Clearly, views on V&V, how they should be accomplished, and under which conditions can
a model be deemed valid are diverse. Different methodologies and perspectives exist on the
subject and are presented in this section.
Carson [1] provides a simple framework for validation of production plant simulation models by introducing practical techniques and guidelines, and categorisation of modelling errors.
The presented framework can be summarised in three consecutive steps as, testing the simulated results for face validity (i.e. if they are reasonable), testing the simulation over a range
of input parameters and finally comparing the simulated results with the reference results.
Another approach to the subject is to use the conserved quantities throughout the system
during specific times or modes of operation for validation purposes.[12] This method analyses
models to understand how conserved quantities (momentum, heat, kinetic energy, etc.) flow
through the model and these quantities are properly conserved.
Sargent, suggests four possible approaches to the management and planning of V&V efforts
and two different paradigms that relate V&V to the model development processes.[2] In this
work, various validation techniques are defined and different aspects of validation, namely conceptual model validity, model verification, operational validity and data validity are explained.
According to Sargent, a model should be developed for a specific purpose (or application) and
its validity should be determined with respect to that purpose. A model is considered valid
for a set of experimental conditions if the models accuracy is within its acceptable range and
the accuracy requirement should be defined at the start of the development project. However,
even if a model passes every experimental scenario satisfactorily, there is no guarantee that it
is valid everywhere inside the domain of application.
The four approaches to V&V management according to Sargent are subjective decision of
the model development team, subjective decision of the model user (customer), independent
V&V where a third party runs the V&V work, and scoring where subjectively determined
scores for various aspects of the simulation models performance are assigned.
A similar scoring approach to increase the credibility of simulation models is developed by
National Aeronautics and Space Administration (NASA).[13] The approach categorises V&V
stages and assesses the exerted effort according to a rigour scale for each of the categories.
This creates an easy to handle overview of the V&V work for the decision-makers.
Oberkampf and Barone [14] attempted to devise a methodology to construct validation
metrics. According to this study, a validation metric should be quantitative, should include
any error resulting from measurements and post processing of experimental data and numerical
operations, should depend on the number of experimental measurements used for testing, and
should exclude any indications of the level of adequacy in agreement between the simulation
and the reality (such as good, excellent or poor).
For the V&V of simulation models, seven rules for model selection and implementation
are proposed by Babuska and Oden [3] for finite element simulation models. This approach
attacks the root of the validation problem, by first selecting an appropriate, well-proposed
189
mathematical model and then selecting the quantities of interest, statistical tolerances for
acceptance accordingly. On the second level, an iterative step takes place where the initial
findings are used to modify and enhance the model. According to this paradigm, verification is performed independent of the validation. Mathematical proof of convergence and the
reproducibility of the experimental results are the final two key concepts that are needed for
a healthy validation effort.
Another methodology for finite element simulation models is A-B-C-D Method, which
defines levels of V&V and approaches the problem from a cost-risk analysis aspect.[9] Here,
A stands for planning, B stands for solution verification, C stands for model validation and
D stands for model validation extrapolated out of the intended scope of application. This
approach introduces a scoring system for different levels of validation, acknowledging that
100% validation is impossible, and the level of attained validation is dependent on the scope
of the application. The needed level of validity comes with a cost to attain it, and this cost is
analysed depending on the application. Also, it has been noted during V&V analysis, that it
is better to use more than one method simultaneously instead of using one optimal method,
since every method has weaknesses and such a practice will remedy these and increase the
model credibility.[15]
In a study by Sarin et al.,[16] a methodology to construct a metric which is used to compare
time histories that are outputs of simulation models with time histories from experimental tests with emphasis on vehicle safety applications is established. The constructed metric
incorporates phase, magnitude and topology features in order to quantify the error between
the simulation and the experiment. Also a regression-based validation model was proposed
which uses this newly developed metric in order to reproduce the subjective judgements of
subject matter experts.
Romero worked on propagating system uncertainties into the simulation model through
model and data conditioning [17] and considers these as an essential step in model
validation.[18] The author considers the combined set of somewhat erred equations and associated compensating parameter values, and looks for effectiveness of the combined set, rather
than correctness of either or both. A model validation activity under representative conditions
is pursued to assess and to hopefully affirm the model, the conclusion being that in any real
validation experiment, there will be some uncertainty in the values of the actual inputs to the
system that is the subject of the model validation inquiry. The logic behind this conclusion
is that validation at the conditions of the validation experiment does not, in general, apply to
where the model will be used because of the different conditions of operation. Model validation and accuracy criteria are almost always substantially subjective and affiliated pass/fail
determinations are not sufficiently robust arbiters of model validity, quality and usefulness.
To extract the most value from validation experiments, any model bias and associated uncertainty should be accounted for in prediction. To accomplish this, a methodology to add the
uncertainty to the model in order to create an augmented or conditioned model that yields total
simulation uncertainty that is compatible with the uncertainty of the conditioned experimental
data was proposed.
Hypothesis testing and Bayesian statistical approaches are also researched as techniques to
validate simulation models. An enhanced Bayesian-based model validation method together
with probabilistic principal component analysis which uses Bayesian hypothesis testing and a
quantitative multivariate validation method based on probabilistic principal component analysis [19] and multivariate Bayesian hypothesis testing are proposed for simulation models
of dynamic systems.[20] These works focus on computer-aided engineering models of automotive safety applications (crash simulation and dummy passenger models), but have the
potential to find usage in vehicle dynamics as well. Using reversed hypothesis testing to validate methods [21] and employing statistical hypothesis testing as a form of objective cost-risk
190
analysis for validation of simulation models [22] are some examples to hypothesis testing
approaches to validation of simulation models. In these approaches the type II error (false
negative), which is the model users risk in modelling practice, is deemed more critical, since
accepting an invalid model as valid will result in the user making analysis with an invalid
simulation model and can lead to damages (even catastrophic results if, for example, the simulation model is for a construction project) and special emphasis is placed on minimising it.
Type I error (false positive) is the model builders risk, since rejecting a valid model will cost
extra work, time and money to the model building party, and does not have the potential to
cause any damage.
3. Validation and vehicle dynamics simulation models

3.1. Vehicle dynamics and modelling
The theory of vehicle dynamics is well established and in this section different sources for
vehicle dynamics and simulation are named.
The earliest and simplest vehicle dynamics model is the single track model.[23] It is still in
use today [24] and it can be traced back to 1940.[25] Fundamentals of modern understanding
of vehicle dynamics and the description of many important characteristics, such as understeer
are presented by Olley [26]. One of the first vehicle models was proposed by Segel [27] for
the time domain analysis, and frequency domain response was explored subsequently in the
1970s by McRuer and Klein.[28]
With the emergence of electronic brake systems and new technologies enabling exertion of
control over many vehicle components in the recent years, complex simulation models have
found a new meaning, thanks to their functional advantages (reproducible results, ability to
simulate inexperimentable situations and fast application) and financial benefits [29] (reduction
of experiment, measurement and prototype costs, early fault detection especially in the cases
when the software of more than one components interacts, better optimisation interface and
faster development cycles).
Many textbooks can be found in the literature which explain the fundamentals [30]
and advanced applications of different aspects of vehicle dynamics such as tyre and
brake dynamics,[23] engine and powertrain management [31] and modelling of vehicle
dynamics.[29,32] Lugner and Plchls work provides an overview of simulation of vehicle
dynamics and model types.[33]
One of the main utilisation fields of vehicle dynamics simulations is driving simulators.
Driving simulators must not only model the dynamics of the vehicle accurately, but also they
must provide correct sensory feedback to the driver. An extensive state of the art survey on
vehicle simulators by Blana can be found in the literature.[34] Another study by Allen et al.
also provides insight on the prospects of the simulator technologies.[35]
Requirements for vehicle dynamics simulation models are explored by Allen and
Rosenthal.[36] Their approach stated that a model must be good enough but not better;
and that the application is what determines the complexity of the model. Ergo, the requirements for any simulation model is application specific. The work emphasises the importance
of an accurate tyre model with appropriate depth for the application, since main phenomenon
causing the dynamics of vehicles, occur between the tyre and the road surface. Tyre modelling
is a fundamental aspect of vehicle handling dynamics and in order to capture the full range of
vehicle stability characteristics, tyre models must include the interaction and saturation characteristics of horizontal slips and camber angle, and properly account for the load variation of
key parameters. Omission of these effects results in a simplified tyre model which excludes
191
roll steer, deflection steer due to compliance and inaccurately calculates individual slip angles
of the tyres.[37] A survey on the modelling applications of the interaction between the tyre
tread-block and the road surface is provided by Wallaschek and Wies.[38]
This importance is further explored in another study with comparisons of tyre models with
different model depths.[39] The effects of different legal tyres on the same vehicle using
fishhook and sine-with-dwell manoeuvres are demonstrated by Arndt et al.[40] Up to 33%
discrepancy is observed for lateral acceleration gain between two approved tyres of the same
manufacturer.
One of the most importance sources on tyre dynamics is written by Pacejka,[41] who
also developed the so-called Magic Formula, an empirical tyre model which relies on curve
fitting using experimentally measured tyre data. Rill [42] developed a first order analytical
tyre model based on Taylor expansion of governing differential equations and another model
based on mechanical analogies is proposed by Lacombe.[43] Analytical models which use
modal parameters are also present.[44] Physical models, which use finite element model to
simulate the mechanics of the tyre structure, are very accurate but need substantial computing
power. Such models are not suitable for online usage. FTire is a recent example to this model
category.[45] Hybrid tyre models combine different model types to exploit their advantages.
A compromise between finite element models and analytical models which is based on the
macroscopic physical description of tyres is proposed by Gallrein and Bcker.[46] Similarly
analytical-empirical hybrid formulations are also existent.[47] Guo and Lu have developed a
semi-physical model, where a detailed nonlinear analytical description of tyre characteristics
is supported with experimentally measured tyre parameters with satisfactory results.[48] The
modelling of tyre wear is also an important aspect. Tyre wear is a major error source in
experimentation. Such models have uses in race performance prediction, tyre development
and fleet management.[49]
3.2. Practice of validation of simulation models for vehicle dynamics
Many of the publications which claim to present a validation methodology or technique tend
to only offer the application of a methodology to an individual case, like vertical dynamics of
articulated vehicles [50] or lateral dynamics of light vehicles.[51] These types of sources
are classified as project specific validation in vehicle dynamics and are explored in this
section together with other relevant research on the subject which do not present a validation
study.
Salaani et al. [52] and Heydinger et al. [53] worked intensively on development, parameter
measurement and validation of vehicle simulation models. A multibody full vehicle model
is developed, parameters for spring, damper, tyre and roll characteristics are measured and
curve fits are generated using these measurements. The performed evaluation covers vehicle
directional dynamics that include steady-state, transient and frequency domain responses. It
is concluded that, any detected discrepancy can be caused by a number of reasons including
model formulation, programming, parameter identification and experimental procedures; and
that the comparison analysis should be supported with analytical reasoning and common sense,
which is a subjective approach.
The methodology consists of three main phases: experimental field data collection, independent vehicle parameter measurement and model formulation, and comparison of simulation
predictions with field data using the same driver control inputs. The importance of independent parameter measurement is emphasised. The model parameters should not be adjusted
according to field tests to obtain a match. The comparisons are performed in time domain to
check the steady state and low-frequency responses and nonlinear effects; and in frequency
domain to check the high-frequency dynamics during transient manoeuvres.
192
The manoeuvres are so sequenced; first quasi-steady state, then step response, then pulse
response (evaluated in frequency domain) and finally a purpose-dependent real-world like
manoeuvre (lane change in this case) are performed. Confidence intervals are constructed,
but no validation criteria are defined using these confidence intervals. Furthermore, no validation metrics are constructed and the validation judgement is taken based on subjective
assessments with no quantitative foundations emphasising the adequacy of the simulation
model.
Validity analysis can also be employed for evaluation of identified vehicle parameters. The
application of genetic algorithm to the physical parameter estimation of a multibody vehicle
model for ride analysis is demonstrated in a project by Alasty and Ramezani [54] In this work,
the reference data are obtained using a more complicated multibody model. No metrics or
statistical analysis is utilised and the validation analysis is executed in time domain, although
the simulation model is developed for vehicle ride analysis. This conflict demonstrates the
importance of the planning and analysis of the simulation goals.
In another study by Mcnaull et al.,[55] a heavy truck simulation model is first modified
according to comparison of experimental and simulation results for lateral steady state manoeuvres; and then validated for dynamic response using a transient manoeuvre. The work does
not introduce or explain the methodology but rather is a demonstration that the end result
of the project is successful. Visual graphical comparison technique is used for validation,
but instead of overlaying the graphs, side-by-side placed diagrams are used, which diminish
the credibility of the validation judgement. Also, no metrics or statistical analysis is performed. The study demonstrates the correct way of using experimental data to correct the
simulation model, by determining the steady state offset and then testing the modified system with a transient manoeuvre. On the other hand applied validation technique, side-by-side
representation of quantities of interests, somewhat lowers the possibility of a healthy call for
validity.
Wade-Allen et al. discuss the validation of a full vehicle model in their 2002 paper.[56]
The research points out the importance of performing the parameter measurements in the
targeted operating regime. If the vehicle model is aimed for simulation of limit handling
scenarios, such as roll over or tyre saturation, the parameter measurements of the subsystems
of the simulation model must be accordingly measured, such as the tyre data over large slip
conditions and higher than normal load, and other nonlinearities due to larger deflections
caused by the highly dynamical manoeuvres.
The addressed validation issues include model formulation, verification of the computer coding, appropriate parameter estimation and measurement procedures and comparison between
experimental and simulation results. It is noted that a thorough validation analysis should
include both steady state and transient manoeuvres, evaluated in both time and frequency
domains. The model is tested using quasi-steady state steering wheel ramp input, pulse
response (in frequency domain), double lane change and fishhook manoeuvres. Validation
metrics or confidence intervals are not used and no statistical analysis is performed. A subjective and qualitative judgement is reached through visual graphical comparison of overlaid
time histories of test and simulation results.
This work reflects a correct approach to the validation problem, but with several shortcomings. The importance of parameter estimation and data validity is well emphasised, and
the sequencing of test manoeuvres, from steady state to transient and to real-life imitating
manoeuvres is proper. An alternative manoeuvre selection for frequency response can be sine
sweep, which will have the same power throughout the selected frequency range, contrary to
pulse response. However, transient response in time domain is not tested, and no quantitative criteria are set for validation. The validity judgement is taken according to the subjective
assessment of the visual resemblance of test and simulation results.
193
A similar study by Ozan et al. [57] is a typical example of the bountiful usage of the term
Validation Methodology. In this work a correlation methodology of a multibody simulation
model of a commercial vehicle is presented. A three-stage process is proposed. First, the vehicles suspension trimming at static conditions is implemented into the simulation model. The
second step is the extensive quasi-static testing of the kinematic components of the suspension and steering system, and their correlation to those of the simulation model. The last step
is the dynamical testing through linear swept steering manoeuvre and fishhook manoeuvre,
and visual graphical comparison of the time histories of the experiment measurements and
simulation outputs, without any statistical analysis, validation metric or accuracy criteria.
In summary, first the mass and properties, then kinematic modules of the simulation model
and finally the whole system response are checked using graphical representations. The
explained technique is project specific, and the used methodology to pass validation judgement lacks traceability and objectivity. The reasoning in the selection of manoeuvres used
in validation of the systems response is not explicit, and the assessment criteria are vague.
Neither validation metrics nor confidence intervals are constructed.
Another study by Hu [58] demonstrates the development of an analytical half vehicle suspension model for suspension control systems analysis and design. The model is validated
based on a comparison of an actual test vehicles and the models simulated time domain
responses over a particular road event which excites the low-frequency band of ride dynamics.
The model parameters are first fine-tuned using the results from a different experiment. This
practice is not advised in general and is only acceptable, if the data used in tuning and validation are different and independent.[53,59] In this study no validation metrics or statistical
analysis is performed. The results of the validation test are evaluated using visual graphical comparison by overlaying time histories of the experimental and model responses on the
same plot. However, the suspension, due its highly dynamic nature because of the constantly
changing vertical forces and the motion of the unsprung and sprung masses, is a subsystem
that should be analysed in the frequency domain. This work is a good example of an analysis
error and demonstrates why the analysis techniques and validity criteria should be defined and
documented at the start of the development project.
An approach to the validation problem as a multi-objective optimisation exercise is presented by Cassara et al.[60] Because of the large number of degrees of freedom and tuneable
parameters of the targeted simulation model, which is a Tractor-Semitrailer model for ride
and handling analysis, such an approach is proposed. To address this complicated question,
modal analysis is performed first at the component level and then at the subsystem level and
then the frequency response of the vehicle system is inspected. Manoeuvre odometrics are
checked as well, and several ride-related components are analysed in the frequency domain.
Focus of the research lies on the subsystem interaction and the effect of frequency response
modelling accuracy of subcomponents to the total system response considering ride and handling. No criteria are used to quantise the quality of the correlation between the experiments
and simulation results. Conclusions underline the importance of frequency domain agreement
of the subcomponents in isolating the problem zones in the simulation model.
A vehicle model/simulation evaluation tool for armed forces, called Model Post Processor is
developed by Howe et al.[61] The tool is capable of comparing different model structures with
each other or with actual static and dynamic test measurements for assessment and evaluation.
Evaluation of static metrics (mass properties, suspension kinematics, compliance, etc.) is
performed by a consistency check subroutine, and dynamic metrics are checked by another
subroutine through a range of test manoeuvres. Dynamics manoeuvre range encompasses
fundamental tests for longitudinal, lateral and vertical dynamics.
Horiuchi et al. have proposed a model-based validation procedure which evaluates the control and intervention performance for the certification of advanced chassis control systems.[62]
194
The proposed validation procedure handled the evaluation of the static and dynamic properties
in two consecutive steps. Braking stability is treated as a static property and is evaluated by
constrained bifurcation and continuation method, whereas the dynamical characteristics of
the vehicle are evaluated by the optimisation-based worst case scenario method. This work is
a good demonstration of application modelling and validation techniques to advanced chassis
control systems and virtual testing.
3.3. Theory of validation of simulation models for vehicle dynamics

The research on methodologies for validation of vehicle dynamics is not diverse. A literature
survey performed by Hoskins and El-Gindhy [63] provides an overview of the validation
methodology studies for vehicle dynamics models used for driving simulators.
One of the most important works on the subject is the 1990 paper of Heydinger et al. [59]
which is arguably the first study to describe a validation methodology for vehicle dynamics
simulation models.
According to this reference, validation is defined as showing that, within some specified
operating range of the vehicle, a simulations predictions of a vehicles responses agree with the
actual measured vehicles responses to within some specified level of accuracy. This definition
emphasises three points:
A simulations predictions may only be correct within some portion of the systems operating
range (e.g. a lateral acceleration range or a steering angle input frequency interval).
A simulations validity is determined for a specified group of inputs and outputs (e.g. a
validated lateral dynamics model with suspension degree of freedom is not necessarily
valid for comfort studies).
A simulations validity is determined according to the variance between the simulations
outputs and experimental measurements.
The described method uses repeated experimental runs at each test condition to generate
sufficient data for statistical analysis and generation of confidence intervals to account for
the random error in the experiments, in both time and frequency domains. Qualitative and
quantitative methods for the comparison of the simulation predictions with the actual test
measurements are considered, and visual graphical comparison method is used.
Another method by Garrott et al. [64] carries on this approach and reapproves the conclusion
that a complete validation analysis should be performed in time and frequency domains. In
this work, six manoeuvre classes are identified and tested. Five of these are identified to be
the primary validation manoeuvres. These are steady state lateral performance (low-frequency
cornering), transient lateral performance (manoeuvres with a broad range of frequencies at the
steering wheel as input), longitudinal acceleration (response to throttle inputs), longitudinal
deceleration (response to braking inputs) and road disturbance input manoeuvres (suspension
kinematics and ride dynamics). The sixth group of manoeuvres, designated as other manoeuvres that attempt to imitate real-life situations (double lane change, fishhook, etc.) are not
considered among the primary validation manoeuvres. Contrary to the preceding study,[59]
no validation metrics or accuracy criteria are used in this work.
Another approach to the problem is suggested by Bernard and Clover.[65] Three questions
are stated to define the validation of a model:
Conceptual validity: Is the model appropriate for the vehicle and manoeuvre of interest?
Verification: Is the simulation based on equations that fully replicate the model?
Data validity: Are the input parameters reasonable?
195
It is argued that due to the increasing complexity of modelling practices, it is generally not
possible to check all the equations (especially in multibody models) and numerical steps, and
running the simulation is the only way for verification.
This method proposes different validation approaches for different model depths. Closed
form solutions or estimates and the lateral load transfer measurements are compared
with the simulation results for manoeuvres lower than 0.5 g which do not involve brake
forces. This approach helps finding errors in inertial and geometric parameters, suspension stiffness concerning handling (cornering, aligning, steering, roll, etc.) and load transfer
model.
For higher than 0.5 g manoeuvres and manoeuvres with tyre saturation (limit handling),
checking the tyre forces as a function of kinematics and normal load is advised. This helps
detecting the errors in tyre model and suspension kinematics.
If the target application for the simulation model involves braking scenarios, checking
longitudinal load transfer, wheel slips, longitudinal tyre forces and, in the case of braking in
a turn manoeuvre, lateral tyre forces assist in finding the errors in longitudinal load transfer,
brake and tyre models.
This work criticises Heydinger et al. [59] for not only increasing confidence in the model,
but also accepting errors as long as the scatter is in an acceptable range, which would mask
the errors that stay within the defined interval. This view is supported with an example case:
an incorrect centre of gravity height measurement, naturally depending on the amount of
error, may provide sufficient results with respect to the confidence intervals for yaw rate and
lateral acceleration; but can be clearly detected by checking the lateral load transfer. On the
other hand, if a manoeuvre in which the lateral load transfer plays a significant role is the
target of the simulation project, such as fishhook manoeuvre, centre of gravity height, roll
angle and lateral load transfer states must be listed among the validation metrics at the start
of simulation project. This critic, therefore, should be directed to not the last stage of the
validation procedure, but to the planning stage, where the target manoeuvre is analysed and
test manoeuvres and validation metrics are chosen.
Another concern that could yield unreliable simulation results is the fact that the road friction
coefficient value supplied to the simulation is most of the time not the same value tested on the
actual test field. Determining or calibrating this value using data from the test vehicle taken
on the test field or directly implementing the manufacturer supplied values can lead to masked
errors.
Concerning the data validity, it is pointed out that faulty data entry is an important risk
factor, possible after the reliability of parameter measurements. According to Bernard and
Clover,[65] the most dangerous part in data entry of parameter values is tyre and suspension
data. Both the tyre model and the suspension model have many parameters, and this step is
prone to human error.
In a follow-up study by Gruening and Bernard,[66] data validity and faulty data entry problem is further investigated and some examples on the effects of different cases of faulty data
entry are demonstrated, although no general methodology to catch such errors is introduced. It
is suggested that unreasonable parameters may arise from three sources; erroneous measurements or bad guesses, misinterpretation of the parameters to be measured or mistakes in data
entry. Other than obvious recommendation of paying extra attention to data entry and checking for mistakes; a pre-processing procedure is suggested. Running the simulation through a
recipe of manoeuvres to determine metrics routinely associated with vehicle performance can
show some of the simple parameter errors, especially those associated with trim conditions
and steady state manoeuvres.
For example, vehicles generally have zero degrees of roll deflection at trim conditions. If a
simulation is run with straight driving at constant speed on a zero friction surface (thus, trim
196
condition), and roll angle is not zero, of course assuming that the mathematical equations of
the model are correct, that indicates that at least one parameter that affects static roll deflection
is wrong. However, even in this simple case, there would be more than one likely cause, for
example one of the parameters associated with tyre geometry or stiffness, or one of the spring
rates.
On the other hand, what if parameter data of one of spring rates and one of the tyres are
mistakenly entered at the same time, in such a way that their effects at trim condition cancel out
each other? In this case, a more dynamical manoeuvre (e.g. steady state cornering), individual
load or force measurements for the tyres, or an isolated test case would be more practical.
In the first two of these suggestions, there is absolutely no guarantee that the simultaneously
erroneously entered (or measured) parameters can be identified. Concerning the third suggestion, generally speaking, it is impossible to devise a manoeuvre which would isolate every
parameter of the system since most of the parameters are inherently interacting. One can only
come up with a limited number of such manoeuvres (for example lateral and longitudinal
manoeuvres can be separated, but the vertical dynamics almost always affect the other two)
but as previously said, there is no guarantee such an error can be detected. Nevertheless, this
approach is very useful in increasing the model confidence.
Allen et al. [67] provided a methodical approach to the validation problem. Possible problem areas causing inconsistencies between computer models and real world are
described as:
Mathematical model.
Computational model programming.
Parameter data.
Numerical accuracy and stability.
It is advocated that the vehicle dynamics model validation must be considered in context
and defined in terms of the domain of useful application, since a simulation model can only
be valid up to a degree and a model should be aimed for a certain behaviour, and a valid
model according to analysis of general system response does not guarantee valid subsystems
models.
The presented validation method is summarised in four steps:
Conceptual validity of the mathematical model.

Face validity (reasonableness) of the simulation model response.
Consistency of input, intermediate and output variables.
Agreement between the simulated behaviour and the reference system (real
or simulated).
For the validation in the lateral direction, three test cases, steady state cornering, sinusoidal
sweep and lane change manoeuvre, are chosen. The research does not offer a way to assess
the findings. Definition of validation metrics, application of statistical methods, or validity
criteria are not discussed.
Of these three approaches to validation methods for vehicle dynamics simulation models,
Heydinger et al. [59] and Garrott et al. [64] focus on operational validity and comparison of
test measurements and simulation results; Bernard and Clover [65] and Gruening and Bernard
[66] recommend analytical solutions and face validity checks for validation and vehicle tests
only for parameter identification and error hunt; and Allen et al. [67] emphasise importance
of face validity, analytical solutions and common sense checks with less methodical approach
to vehicle testing. A summary of these studies is presented in Table 1.

Table 1.
Summary of the included model validation studies.
Reference
Aim of the model
Salaani et al. [52]
Handling
Heydinger et al. [53]
Handling
Alasty and Ramezani [54]
Ride
Mcnaull et al. [55]
Handling
Wade-Allen et al. [56]
Handling
Ozan et al. [57]
Handling
Hu [58]
Ride
Cassara et al. [60]
Ride and handling
Heydinger et al. [59]
Handling
Garrott et al. [64]
Longitudinal,
handling and ride
Bernard and Clover [65]
Handling
Allen et al. [67]
Handling
4.
197
Method
Independent parameter measurement and real reference
data
Steady state, transient, frequency and real life like
manoeuvres
Analysis in time and frequency domains
Confidence intervals and no validation metrics
data
Steady state, step, frequency and real life like
manoeuvres
Confidence intervals and no validation metrics
Reference data from a more complicated model
Analysis in time domain
No statistical analysis and no validation metrics
Model correction using experimental data
Transient manoeuvres
Parameter measurements at targeted operating regime
Steady state and transient manoeuvres
data
Transient manoeuvres
Model correction using experimental data
Modal analysis at subsystem level
Analysis in frequency domain
data
Steady state, transient, frequency and real life like
manoeuvres
Statistical analysis and validation metrics
Manoeuvre classification
Data validity and data entry errors
No manoeuvre selection
Data validity and data entry errors
Steady state, transient and real life like manoeuvres
Conclusion
In this work, a literature survey on approaches to validation of vehicle dynamics simulation

models is presented. Several studies on validation and verification theory are discussed and
general approaches to the validation problem from other fields of engineering are examined.
198
Simulation and validation practices and methodologies in the field of vehicle dynamics are
presented.
There are many and similar definitions for V&V in the literature. One thing nearly all
experts agree upon is that an absolute validation is not possible, and validation analysis should
be handled according to the specific requirements and limitations of the application.
Existent works on validation methodologies for vehicle dynamics simulations focus on
different aspects of the question. Neither there is a standard in experimentation and data
handling processes in vehicle dynamics modelling, nor is there a standard reasoning process
in the vehicle dynamics modelling application in validation analysis. Most of the applications
rely only on visual comparison and subjective judgement. Diagrams types used in visual
comparison also do not follow any recognisable pattern and their contents and structures are
determined at will by the research team. Most of the time, the team which developed the model
also decides if the simulation is valid. This whole process chain diminishes the credibility of
these models.
A vehicle model should be analysed in time and frequency domain using both steady state
and transient manoeuvres. Both analyses can show characteristics which may go undetected
if only one is used. There are basic manoeuvres which demonstrate the general dynamics of
the vehicle, and then, there are other manoeuvres, which imitate real-life scenarios.
The validation criteria are dependent on the application; the validity metrics and the data
handling are dependent on the chosen manoeuvres and analysis; and the manoeuvre selection
is dependent on the targeted real-life phenomena to be simulated. Thus, validation of a vehicle
dynamics simulation model should be planned according to the characteristics of the real-life
phenomena it is trying to reproduce. The test manoeuvres to be used in the validation should
be determined early in the project timeline and according to the analysis of the real event to
be simulated.
To this aim a validation recipe, so to say, which incorporates a top-down approach is required.
Real-world manoeuvres and test manoeuvres need to be analysed, and the relationship in
between needs to be examined. For each test manoeuvre, experimental procedures, data handling techniques, validation metrics and accuracy criteria should be explicitly declared in
the project documentation. Such an approach will help to achieve more time and cost efficient simulation projects with increased model confidence by enhancing the traceability of the
validation process.
References
[1] Carson JS. Model verification and validation. Winter Simulation Conference; San Diego, CA; 2002.
[2] Sargent RG. Verification and validation of simulation models. Winter Simulation Conference; Austin, TX; 2010.
[3] Babuska I, Oden JT. Verification and validation in computational engineering and science: basic concepts.
Comput Meth Appl Mech Eng. 2004;193:40574066.
[4] American Institute of Aeronautics and Astronautics. Guide for the Verification and validation of computational
fluid dynamics simulations; AIAA; 1998. (G-077-1998e).
[5] Stern F, Wilson RV, Coleman HW, Paterson EG. Comprehensive approach to verification and validation of CFD
simulations part 1: methodology and procedure. J Fluids Eng. 2001;123:793801.
[6] Schlesinger S, Buyan JR, Callender ED, Clarkson WK, Perkins FM. Developing standard procedures for
simulation validation and verification. Summer Computer Simulation Conference; Houston, TX; 1974.
[7] Schlesinger S. Terminology for model credibility. Simulation. 1979;32(3):103104.
[8] Law AM. Simulation modeling & analysis. 4th ed. New York: McGraw-Hill; 2007.
[9] Logan RW, Nitta CK. Verification & validation: process and levels leading to qualitative or quantitative validation
statements. Warrendale (PA): SAE International; 2004. (SAE 2004-01-1752).
[10] Popper K. Logik der Forschung. 10th ed. Tbingen: Mohr Siebeck; 2005.
[11] Klein EE, Herskovitz PJ. Philosophical foundations of computer simulation validation. Simulat Gaming.
2005;36(3):303329.
[12] Tiller MM. Verification and validation of physical plant models. Warrendale (PA): SAE International; 2009.
(SAE 2009-01-0527).
199
[13] Blattnig SR, Green LL, Luckring JM, Morrison JH, Tripathi RK, Zang TA. Towards a credibility assessment of
models and simulations. 10th AIAA Non-Deterministic Approaches Conference; Schaumburg, IL; 2008.
[14] Oberkampf WL, Barone MF. Measures of agreement between computation and experiment: validation metrics.
J Comput Phys. 2006;217:536.
[15] Logan RW, Nitta CK. Comparing 10 methods for solution verification, and linking to model validation. J Aerosp
Comput Inf Commun. 2006;3:354373.
[16] Sarin H, Kokkolaras M, Hulbert G, Papalambros P, Barbat S, Yang RJ. A comprehensive metric for comparing
time histories in validation of simulation models with emphasis on vehicle safety applications. ASME International Design Engineering Technical Conferences and Computers and Information in Engineering Conference;
New York City, NY; 2008.
[17] Romero VJ. Type X and Y errors and data & model conditioning for systematic uncertainty in model calibration.
Warrendale (PA): SAE International; 2008. (SAE 2008-01-1368).
[18] Romero VJ. Validated model? Not so fast. The need for model conditioning as an essential addendum to model
validation. 48th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference;
Honolulu, HI; 2007.
[19] Fu Y, Zhan Z, Yang R. A study of model validation method for dynamic systems. Warrendale (PA): SAE
International; 2010. (SAE 2010-01-0419).
[20] Jiang X, Yang R, Barbat S, Weerappuli P. Bayesian probabilistic PCA approach for model validation of dynamic
systems. Warrendale (PA): SAE International; 2009. (SAE 2009-01-1404).
[21] Hartmann C, Smeyers-Verbeke J, Pennickx W, Heyden YV, Vankeerberghen P, Massart DL. Reappraisal of
hypothesis testing for method validation: detection of systematic error by comparing the means of two methods
or of two laboratories. Anal Chem. 1995;67(24):44914499.
[22] Balc S, Sargent RG. Some examples of simulation model validation using hypothesis testing. Winter Simulation
Conference; San Diego, CA; 1982.
[23] Ammon D. Modellbildung und Systementwicklung in der Fahrzeugdynamik. Stuttgart: B. G. Teubner; 1997.
[24] Diebold L, Schindler W, Haug J, Daesch C, Lahti M. Einspurmodell fr die Fahrdynamiksimulation und -analyse.
ATZ. 2006;11(108): 962967.
[25] Riekert P, Schunk TE. Zur Fahrmechanik des gummibereiften Kraftfahrzeugs. Ingenieur-Archiv. 1940;11:
210224.
[26] Olley M. Road manners of the modern car. Proc Inst Automobile Eng. 1946;41(1):523551.
[27] Segel L. Theoretical prediction and response of the automobile to steering control, research in automobile
stability and control in tire performance. Proc Automobile Div Inst Mech Eng. 1956;7:2646.
[28] McRuer D, Klein R. Automobile controllability driver/vehicle response for steering control volume 1
summary report. Warrendale (PA): SAE International; 1975. (NHTSA DOT HS-901 407).
[29] Schramm D, Hiller M, Bardini R. Modellbildung und simulation der dynamik von Kraftfahrzeugen. Berlin,
Heidelberg: Springer; 2010.
[30] Zomotor A. Fahrwerktechnik: Fahrverhalten. 2nd ed. Wrzburg: Vogel; 1991.
[31] Kiencke U, Nielsen L. Automotive control systems for engine, driveline and vehicle. 2nd ed. NewYork: Springer;
2007.
[32] Genta G. Motor vehicle dynamics. Singapore: World Scientific; 1997.
[33] Lugner P, Plchl M. Modelling in vehicle dynamics of automobiles. Z Angew Math Mech. 2004;84(4):219236.
[34] Blana E. A survey of driving research simulators around the world, institute of transport studies. Working Paper
481. University of Leeds; 1996 [cited 2013 Sep 12]. Available from: http://eprints.whiterose.ac.uk/2110/.
[35] Allen RW, Fancher PS Jr, Levison WH, Machey J, Mourant RR, Schnell T, Srinivasan R. Simulation and
measurement of driver and vehicle performance. Transportation in the New Millenium; 2000 [cited 2013 Sep
12]. Available from: http://trid.trb.org/view.aspx?id=639809.
[36] Allen RW, Rosenthal TJ. Requirements for vehicle dynamics simulation models. Warrendale (PA): SAE
International; 1994. (SAE 940175).
[37] Bundorf RT, Leffert RL. The cornering compliance concept for description of vehicle directional control
properties. Warrendale (PA): SAE International; 1976. (SAE 760713).
[38] Wallaschek J, Wies B. Tyre tread-block friction: modelling, simulation and experimental validation. Veh Syst
Dyn. 2013;51(7):10171026.
[39] Allen RW, Magdaleno RE, Rosenthal TJ, Klyde DH, Hogue JR. Tire modeling requirements for vehicle dynamics
simulation. Warrendale (PA): SAE International; 1995. (SAE 950312).
[40] Arndt MW, Rosenfield M, Arndt SM. How tires change a SUVs performance in fishhook and sine-with-dwell
testing. 21st International Technical Conference on the Enhanced Safety of the Vehicles; Stuttgart; 2009.
[41] Pacejka HB. Tyre and vehicle dynamics. 2nd ed. Oxford: Butterworth Heinemann; 2005.
[42] Rill G. First order tire dynamics. III European Conference on Computational Mechanics, Solids, Structures and
Coupled Problems in Engineering; Lisbon; 2006.
[43] Lacombe J. Tire model for simulations of vehicle motion on high and low friction road surfaces. Winter
Simulation Conference; Orlando, FL; 2000.
[44] Dihua G, Zhaolong D. Chengjian F. Tire model by using modal parameters directly (MPTM). Warrendale (PA):
SAE International; 2007. (SAE 2007-01-1512).
[45] Gipser M. FTire the tire simulation model for all applications related to vehicle dynamics. Veh Syst Dyn.
2007;45: 139151.
200
[46] Gallrein A, Bcker M. CDTire: a tire model for comfort and durability applications. Veh Syst Dyn.
2007;45(Supp.1):6977.
[47] Gim G, Choi Y, Kim S. A semi-physical tire model for a vehicle dynamics analysis of handling and braking.
Veh Syst Dyn. 2007;45(Supp.1):169190.
[48] Guo K, Lu D. UniTire: unified tire model for vehicle dynamic simulation. Veh Syst Dyn. 2007;45(Supp.1):7999.
[49] Braghin F, Cheli F, Melzi S, Resta F. Tyre wear model: validation and sensitivity analysis. Meccanica.
2006;41:143156.
[50] Cole DJ, Cebon D. Validation of an articulated vehicle simulation. Veh Syst Dyn. 1992;21(1):197223.
[51] Nalecz AG, Lu Z, DEntremont KL. Development and experimental validation of advanced dynamic vehicle
simulation (ADVS). Veh Syst Dyn. 1994;23(Supp.1): 390410.
[52] Salaani MK, Schwarz C, Heydinger GJ, Grygier PA. Parameter determination and vehicle dynamics modeling
for the national advanced driving simulator of the 2006 BMW 330i. Warrendale (PA): SAE International; 2007.
(SAE 2007-01-0818).
[53] Heydinger GJ, Schwarz C, Salaani MK, Grygier PA. Model validation of the 2006 BMW 330i for the national
advanced driving simulator. Warrendale (PA): SAE International; 2007. (SAE 2007-01-0817).
[54] Alasty A, Ramezani A. Genetic algorithm based parameter identification of a nonlinear full vehicle ride model.
[55] Mcnaull PJ, Guenther DA, Heydinger GJ, Grygier PA, Salanni MK. Validation and enhancement of a heavy
truck simulation model with an electronic stability control model. Warrendale (PA): SAE International; 2010.
(SAE 2010-01-0104).
[56] Wade-Allen R, Chrstos J, Howe G, Klyde DH, Rosenthal TJ. Validation of a non-linear vehicle dynamics
simulation for limit handling. Proc Inst Mech Eng D J Auto Eng. 2002;216:319327.
[57] Ozan B, sendur P, Uyank ME, z Y, Ylmaz SI. A model validation methodology for evaluating rollover
resistance performance of a ford commercial vehicle. Warrendale (PA): SAE International; 2010. (SAE-201001-0107).
[58] Hu H. Experimental validation of a half-vehicle suspension model. Warrendale (PA): SAE International; 1993.
(SAE 931966).
[59] Heydinger GJ, Garrott WR, Chrstos JP, Guenther DA. A methodology for validating vehicle dynamics
simulations. Warrendale (PA): SAE International; 1990. (SAE-900128).
[60] Cassara SJ, Anderson DC, Olofsson JM. A multi-level approach for the validation of a tractor-semitrailer ride
and handling model. Warrendale (PA): SAE International; 2004. (SAE 2004-04-2694).
[61] Howe JG, Chrstos JP, Romano R, OKins J. Development of a vehicle model/simulation evaluation tool.
[62] Horiuchi S, Okada K, Nohtomi S. Model-based validation procedure for the certification of advanced chassis
control systems. Veh Syst Dyn. 2010;48(Supp.1):393409.
[63] Hoskins AH, El-Gindhy M. Technical report: literature survey on driving simulator validation studies. Int J
Heavy Veh Syst. 2006;13(3):241252.
[64] Garrott WR, Grygier PA, Chrstos JP, Heydinger GJ, Salaani MK, Howe JG, Guenther DA. Methodology for
validating the national advanced driving simulators vehicle dynamics (NADSdyna). Warrendale (PA): SAE
[65] Bernard JE, Clover CL. Validation of computer simulations of vehicle dynamics. Warrendale (PA): SAE
[66] Gruening J, Bernard JE. Verification of vehicle parameters for use in computer simulation. Warrendale (PA):
SAE International; 1996. (SAE 960176).
[67] Allen RW, Rosenthal TJ, Klyde DH, Owens KJ, Szostak HT. Validation of ground vehicle computer simulations
developed for dynamics stability analysis. Warrendale (PA): SAE International; 1992. (SAE 920054).
2013 IEEE International Conference on Computer Vision Workshops
Various Approaches for Driver and Driving Behavior Monitoring:

A Review
Hang-Bong Kang
Dept. of Digital Media
Catholic University of Korea
hbkang@catholic.ac.kr
monotonous roads [2]. Several studies have produced
various estimates of the level of sleep deprivation as it
relates to road accidents. In addition, driver distraction or
inattention is another critical problem for safe driving [3].
In summary, driver drowsiness and distraction are major
causal factors behind road accidents.
To reduce the number of road accidents, it is necessary to
monitor driver and driving behavior and alert the driver
when he or she is drowsy or in distraction state. In addition,
if it were possible to predict unsafe driving behaviors in
advance, this would contribute to safe driving. According
to one report [4], the amount of car crashes would be
reduced by 10- 20% by monitoring and predicting driver
and driving behaviors. A reliable and robust driver
drowsiness and distraction detection system would send an
alert to the driver and thus reduce the number of hazardous
situations on the road. If it were possible to predict unsafe
driving behavior in advance, this would also be helpful in
preventing road accidents. Thus, it is desirable to design a
framework consisting of two phases, that is, both
monitoring and predicting driver and driving behavior.
Figure 1 shows such a framework.
For a driver monitoring system, two issues such as driver
fatigue measurement and distraction detection should be
solved. Usually, driver fatigue or drowsiness may be
related with symptoms including eye movement, facial
expression, heart and breathing rate, and brain activity
[5-10]. To detect driver drowsiness, visual features such as
eye movement and facial expression are very important.
Yawning measurement is also good indicator of a drivers
drowsiness [11]. As non-visual features, heart rate
variability (HRV), galvanic skin response (GSR) and
conductivity, steering-wheel grip pressure, and body
temperature are possible candidates for estimating the
drivers fatigue level indirectly [12]. Electroencephalogram
(EEG) and Electro-oculogram (EoG) give additional
psychophysiological information about drowsiness or
emotional reactions [13]. Driving behavior information
such as steering wheel movement, lane keeping,
acceleration pedal movement and braking, etc., should also
be considered to detect driver drowsiness.
Abstract
In recent years, driver drowsiness and distraction have
been important factors in a large number of accidents
because they reduce driver perception level and decision
making capability, which negatively affect the ability to
control the vehicle. One way to reduce these kinds of
accidents would be through monitoring driver and driving
behavior and alerting the driver when they are drowsy or in
a distracted state. In addition, if it were possible to predict
unsafe driving behavior in advance, this would also
contribute to safe driving. In this paper, we will discuss
various monitoring methods for driver and driving
behavior as well as for predicting unsafe driving behaviors.
In respect to measurement methods of driver drowsiness,
we discussed visual and non-visual features of driver
behavior, as well as driving performance behaviors related
to vehicle-based features. Visual feature measurements
such as eye related measurements, yawning detection,
facial expression are discussed in detail. As for non-visual
features, we explore various physiological signals and
possible drowsiness detection methods that use these
signals. As for vehicle-based features, we describe steering
wheel movement and the standard deviation of lateral
position. To detect driver distraction, we describe head
pose and gaze direction methods. To predict unsafe driving
behavior, we explain predicting methods based on facial
expressions and car dynamics. Finally, we discuss several
issues to be tackled for active driver safety systems. They
are 1) hybrid measures for drowsiness detection, 2) driving
context awareness for safe driving, 3) the necessity for
public data sets of simulated and real driving conditions.
1. Introduction
Recently, the total number of serious car crashes is still
increasing regardless of improvements in road and vehicle
design for driver safety. The U.S. National Highway
Traffic Safety Administration (NHTSA) data indicate that
more than 40,000 Americans suffer serious injuries from
56,000 sleep related road crashes annually [1]. According
to a study by the Sleep Research Center (UK), driver
drowsiness at the wheel causes up to 20% of accidents on
978-0-7695-5161-6/13 $31.00 2013 IEEE
978-1-4799-3022-7/13
DOI 10.1109/ICCVW.2013.85
616
Recently, excessive uses of in-vehicle information

systems such as navigation systems and mobile phones
induce visual and cognitive distraction in the driver. Visual
distraction refers to the state of eye-off-road, and
cognitive distraction is described as mind-off-road [14].
Driver distraction may lead to larger lane variation, slower
response to obstacles, and more abrupt steering control, and
its monitoring should be a feature of a safer
driver-monitoring system. To detect driver distraction, it is
necessary to extract head pose or gaze information
efficiently.
By carefully monitoring driver and driving performance
behavior, it is possible to predict minor and major accidents.
In particular, the progress of pervasive computing
technology with integrated sensors and networking has
made it possible to build an ideal platform to predict
accidents. Jabon et al. [15] identify a comprehensive set of
drivers key facial features at various pre-accident intervals
and use them to predict minor and major accidents. This
approach is very important for active driver-safety systems
designed to prevent accidents. The Signal Processing 5
Laboratory of the EPFL in Switzerland [16] analyzes facial
expressions and muscle movements to detect distraction as
well as emotions that could indicate that the driver is not up
to the task at hand.
The aim of this paper is to discuss monitoring the
drivers state as well as predicting unsafe driving behaviors.
We explain several issues in developing a framework
consisting of two phases: monitoring and predicting driver
and driving behavior. The organization of this paper is as
follows: Section 2 discusses driver drowsiness detection.
We discuss driver behavior features such as visual and
non-visual features, and driving performance behaviors
related to vehicle-based features. Eye related measurement
like PERCLOS, yawning detection and some current
limitations in measuring visual features are discussed in
detail. As for non-visual features, we explore physiological
signals for detecting drowsiness. As for vehicle-based
features, we describe steering wheel movement and the
standard deviation of lateral position. Section 3 describes
some issues related to driver distraction measurement, in
particular, head pose and gaze direction methods. Section 4
presents prediction methods for unsafe driving behaviors.
We explain predicting methods based on facial expression
and car dynamics. Finally, Section 5 discusses some issues
for active driver safety systems. They are 1) hybrid
measures for drowsiness detection, 2) driving context
awareness for safe driving, 3) the necessity for public data
sets of simulated and real driving conditions.
Figure 1: General framework of a monitoring and predicting

unsafe behavior system
he or she is drowsy or a state of inattention. In this Section,

we will discuss drowsiness measurement methods.
The word drowsy simply refers to an inclination to fall
asleep. A drowsy driver who falls asleep at the wheel can
be characterized by diminished alertness compared to a
normal state. Sometimes a driver experiences sleep for a
few seconds and may not even realize it. This is called
micro-sleep. The duration of micro-sleep can last between a
few seconds and as long as 30 seconds or even more. This is
sufficient time to drift out from ones traffic lane and crash
into a tree or another car. Therefore, the drivers drowsiness
state, in which a transition occurs from awake to asleep,
should be monitored.
To detect the drowsiness level of the driver, we have to
extract driver behavior information as well as driving
behavior information as shown in Table 1. Driver behavior
information consists of both visual and non-visual features.
Visual features include eye closure, eye blinking, yawning,
head pose, facial expression etc. [10, 11, 17]. The
frequency of eye blinking and degree of eyelid opening are
a good index of the tiredness level [10]. Non-visual features
consist of heart rate, pulse rate and brain activity. These
physiological
signals
(electrocardiogram
(ECG),
electromyogram (EMG), electro-oculogram (EoG) and
electroencephalogram (EEG)) are used to detect driver
drowsiness [18-23]. Driving behavior information includes
deviations from lane position, vehicle speed, steering
movement, pressure on the acceleration pedal, etc. [24, 25].
2.1. Visual features

Usually, facial movements such as eye blinking, frequent
yawning and nodding or swinging head are key elements
among visual features used for detecting drowsiness. Much
research work is focused on eye behaviors in particular to
determine a drivers alertness [26, 27]. A reliable and valid
determination of a drivers alertness level is known as
PERCLOS (Percent Eye Closure) [28-30]. PERCLOS is
2. Driver drowsiness measurement

For safe driving, it is necessary to construct a reliable
driver monitoring system which could alert the driver when
617
Table 1: Summary of various features for detecting and predicting driver drowsiness.
on the brows, mouth and nasolabial fold are good physical

signs that drowsiness is being resisted, and that therefore it
is present [38].
the percentage of total time that the drivers eyelid is closed

80% (or more) over the pupil and also reflects slow eyelid
closure. When PERCLOS exceeds a predetermined
threshold, the proposed system generates a drowsiness
warning. However, one disadvantage of PERCLOS is that
sometimes a driver who is trying to stay awake is able to
fall asleep with his eyes open.
To calculate PERCLOS, we have to extract the eye
region including the pupil area. However, there are some
limitations in extracting those visual features. One of them
is the problem of proper lighting. Drowsiness should be
detected under real conditions, i.e., throughout daytime and
night, and regardless of whether the driver is wearing
glasses or sunglasses. Usually, a simple CCD or web
camera is used during the day, and an IR camera is used at
night [31-34]. Moreover, for eye detection with a driver
who is wearing sunglasses, it is necessary to find a proper
wavelength of Near IR (NIR) illumination. One possible
candidate wavelength is 850nm. In a real automotive
environment, reflected sunlight is also generated on the
outer surface of the eyeglasses. To diminish the reflection
effect, Jo et al. [34] used a NIR illuminator with a narrow
bandpass filter which restricts the incoming wavelength of
light to 850nm. This is because the high-power LED
illuminator is more powerful than the sunlight in the car.
Figure 2 shows that the reflected sunlight is removed by an
NIR illuminator with narrow bandpass filter [34].
Another method is faceLAB used in the commercial
product such as the Seeing Machine [35]. A passive pair of
video cameras is used and the video images are processed
in real-time to determine the 3D position of each feature.
The system is able to determine a precise 3D head pose and
computes eye gaze direction. Its advantages include coping
well with low light conditions and head movements while
the driver is wearing sunglasses.
Yawning is another sign of driver drowsiness. This is
detected from measuring both the rate and the amount of
changes in the drivers mouth contour [7, 11]. Head pose
estimation and head motion detection of movements such
as nodding are also important in monitoring driver alertness
[36, 37]. In addition, facial wrinkles of the driver appearing
2.2. Non-visual features

Non-visual features or physiological signals such as
heart rate and brain activity are useful in predicting
drowsiness, with fewer false positives compared to visual
features because the determination of a drowsy state from
visual features can be possible only after the driver is well
on the way to sleep. In other words, the prediction of
drowsiness based on these physiological signals makes it
possible to warn a drowsy driver in a timely manner.
Electrocardiogram (ECG), electroencephalogram (EEG),
electromyogram (EMG), electro-oculogram (EoG), and
Photoplethysmography (PPG) may all be used as
physiological signals.
Figure 2: Reflected sunlight is dealt with by an NIR illuminator

with a narrow bandpass filter in [34]. (a)NIR illuminator only,
(b) NIR illuminator with bandpass filter.
618
From the ECG signal, heart rate (HR) can be extracted;

the heart rate can be used to detect drowsiness because it
varies significantly between alertness and drowsiness states
[19, 39]. Heart rate variability (HRV) which measures the
beat-to-beat changes in the heart rate is also used to detect
drowsiness. As the driver goes from an alert to a drowsy
state, the ratio of low frequency to high frequency beats in
the ECG signal progressively decreases [20, 40].
One critical issue in handling physiological signals is to
eliminate noise and artifacts inevitable in real environment
driving conditions. Following effective filtering, various
feature extraction techniques such as Fast Fourier
Transform (FFT) and Discrete Wavelet Transform (DWT)
are used. Then, the extracted features are classified using
Support Vector Machine (SVM), Artificial Neural
Networks (ANN), Linear Discriminant Analysis (LDA),
etc. [40-43].
Even though the reliability and accuracy in detecting
drivers drowsiness based on physiological signals is high
when compared to visible features, an important limitation
of physiological signal measurement is its intrusive nature.
One possible way to solve this limitation is to use wireless
technologies such as Zigbee and Blutooth for measuring
physiological signals in a non-intrusive way by placing the
electrodes on the steering wheel or in the drivers seat [44,
45]. Figure 3 shows ECG measurement on the driver
seatback [45]. Finally, the signals are handled by smart
phones and driver drowsiness is determined [46]. However,
this kind of non-intrusive system is less accurate compared
to intrusive systems due to improper electrode contact.
To obtain reliable drivers drowsiness detection results,
some attempts had been done to fuse various measurements
[20, 47]. A mixture of PERCLOS, ECC and EEG were used
to detect driver drowsiness and resulted in a higher success
rate than individual measures [20]. Cheng et al. [47] used
fusion of PERCLOS, blink rate, maximum close duration
and percentage of non-steering measures to detect
drowsiness.
Figure 3: ECG measurements on driver's seatback in [45].
drowsiness state is determined from small SWMs of

between 0.5 and 5. SWMs are being adopted by car
companies such as Nissan and Renault, but work in very
limited situations due to their reliability only in particular
environments [23].
Standard Deviation of Lateral Position (SDLP) is
another sleepiness sensitive continuous performance
measure. Ingre et al. [52] found that SDLP is correlated
with the Karolinska Sleepiness Scale (KSS), a nine-point
scale that has verbal anchors for each step. However, SDLP
is dependent on external factors such as road markings,
lighting and climatic conditions. Sometimes, these driving
performance measures are not specific to the drivers
drowsiness. In particular, these kinds of driving behavior
measures are dependent on the vehicle type, driver
experience, and conditions of the road.
3. Driver distraction detection

Distraction is another important factor causing
impairment of driver attention, involving a driver not
paying sufficient attention to the road in spite of the
presence of obstacles or other people. In particular, there is
a trend toward increasing use of in-vehicle information
systems, which also leads to driver distraction.
A good first step in detecting driver distraction or
inattention is to monitor the driver head pose and gaze
direction. A forward warning system [53] uses driver
behavioral information to determine driver distraction and
to determine whether the driver is looking straight ahead.
Murphy-Chutorian et al. [54] use head pose information
extracted from a localized gradient histogram and support
vector regressors (SVRs) to recognize driver awareness.
Kaminski et al. [55] propose a system to estimate
continuous head orientation and gaze direction.
Distractions can be categorized as visual, that is,
eye-off-road, and cognitive, that is, mind-off-road [17].
Liang [3] proposes a method to detect the interactions of
visual and cognitive distractions.
To detect driver distraction, it is necessary to extract
head pose or gaze information [56, 57]. Head pose
estimation provides a drivers eld of view and current
focus of attention. It is intrinsically linked to visual gaze
direction. When the eyes are not visible, head pose is used
to estimate the gaze direction. The combination of both
2.3. Driving behavior features

Driving behavior features or driving performance
measures include steering wheel movement, lane keeping,
acceleration pedal movement and braking, etc. [48-50].
These features correlate to vehicle type and variability
among drivers in their driving habits, skills and experience.
The two most commonly used driving behavior measures
for detecting the level of driver drowsiness are the steering
wheel movement and the standard deviation in lateral
position.
Steering Wheel Movement (SWM) is measured using
steering angle sensor mounted on the steering column.
When the driver is drowsy, the number of
micro-corrections to the steering wheel, which are
necessary in normal driving, is reduced [51]. The drivers
619
Although facial features have proven to aid in predicting

minor accidents, their predictive efficacy should be
improved. To predict accidents more accurately, it is
necessary to capture other physiological signals either from
the driver or from some other part of the
driver-environment system. Based on these data, a new
model can be generated for predicting impending driver
accidents. In particular, various populations of participants,
traffic cultures and driving contexts should be handled in
constructing a more extensive and general accident
prediction model.
head pose and eye direction provides a persons gaze

information [58].
Murphy-Chutorian et al. [56] categorize eight conceptual
approaches to head pose estimation. These are appearance
template methods, detector array methods, nonlinear
regression methods, manifold embedding methods, flexible
models, geometric methods, tracking methods, and hybrid
methods. Feature-based methods are the most commonly
used gaze direction estimation methods [57]. With these
methods, local features such as contours, eye corners, and
reflections from the eye images are used. However, in
vehicular environments, general gaze direction is
sometimes good enough to reduce false warnings in
forward collision warning (FCW) systems [59]. General
gaze direction can be approximated by using only head
orientation, which is computed by shape features
with/without eye position, texture features or hybrid
features consisting of shape and texture features [60].
5. Discussion
For active driver safety systems, we have discussed
various topics such as driver drowsiness detection, driver
distraction detection, and prediction of unsafe driving
behavior. To develop a better driver safety system, several
issues should be addressed. The most important ones are a)
hybrid measures for drowsiness detection, b) driving
context-awareness for safe driving, c) the necessity for
public data sets for simulation and real driving conditions.
4. Predicting unsafe driving behavior

Monitoring driver state, driving behavior performance
and vehicle state is very important for improving active
driver safety systems. The driver state is monitored by
measuring drowsiness, fatigue or stress levels [61-64].
Driving behavior performance and vehicle state are also
monitored by analyzing the information regarding driving
speed, steering wheel angle, braking, and acceleration
[48-50, 65, 66]. After detecting drowsiness or distraction,
an alert is sent to the driver.
Another important issue for an active driver safety
system is to develop a mechanism to predict minor and
major accidents in advance. Jabon et al. [15] used facial
features to aid in driver accident prediction. They combined
both vehicle dynamics and driver face analysis for accident
prediction. First, a comprehensive set of 22 raw facial
features are analyzed. Then, the most valuable statistics for
predicting accidents are extracted from a range of time and
frequency domain values, so that both major and minor
accidents are can be predicted. Even though the
experimental results of Jabon et al. [15] are not based on
real road situations, it has been found that facial features
show most predictive accuracy four seconds prior to
accidents and are more helpful in predicting minor
accidents than major ones. This is because predictive
accuracy for major accidents comes primarily from vehicle
features rather than facial features.
EPFL and PSA Peugeot Citroen [16] are developing a
technology to detect the driver distraction as well as
emotions that could indicate that the driver is not up to the
task at hand. In other words, facial expressions and muscle
movements are important in analyzing whether the driver is
too distracted, too tired or even too angry to safely control
their vehicle.
5.1. Hybrid measures for drowsiness detection

Among driving behavior and driver behavior features in
detecting driver drowsiness, driving behavior may
sometimes not detect a drivers drowsiness reliably. Driver
behavior features are better than driving behavior features,
but visual features have sometimes limitations due to
illumination conditions and driver posture [34]. Non-visual
features such as physiological features are reliable and
accurate, but their nature is intrusive. This should be solved
before they can be used in real vehicular environments.
Even though a less intrusive measurement of ECG has been
developed [45], EEG and EoG still require electrodes
placed on the scalp or eye area in an intrusive manner.
However, non-intrusive measurement of physiological
signals may be developed in the near future.
For reliable drowsiness detection, a hybrid measure
fusing from visual, physiological and driving behavior
features is desirable. The fusion method usually shows
good performance in drowsiness detection even when
certain sensors lose validity. Figure 4 shows an example of
hybrid measurement for driver state detection. One issue is
to develop a reliable data fusion method at the feature-level
or decision-level [20].
5.2. Driving context-awareness for safe driving

For safe driving, driving context awareness is necessary
because various information related to driving conditions
and environment should be explored effectively. We can
divide driving context into global and local. The global
driving context refers to driving environment parameters
such as vehicle type, road type, driving time, driving
620
circumstances, road conditions, etc. The local driving

context refers to the driver status. In other words, local
context is related to the drivers visual and cognitive
perceptiveness and its deterioration because of distraction,
drowsiness and / or emotions.
In particular, a driving context-based computational
model combining driving environment and driver status
seems to be a promising avenue of development. Using
such a model, the detection rate for distraction and
drowsiness will increase, which will be helpful in
predicting unsafe driving behavior.
5.3. Necessity for public data sets for simulation

and real driving conditions
To monitor driver and driving behavior, various
hardware and software algorithms are being developed, but
they are tested mostly in the simulated environments
instead of real driving ones. This is due to the danger of
testing drowsiness in real driving environments. Philp et al.
[67] found that reaction times and sleepiness as derived
from self-evaluations increase in a simulated environment
compared to a real driving environment due to the
monotony of the experience. Engstorm et al. [68] stated that
physiological workload and steering activity were both
higher under real driving conditions compared to simulated
environments. In real driving conditions, various factors
including variations in lighting and noise can also affect the
drivers attention. Thus, it is necessary to make simulated
environments looks more like realistic.
Even though various kinds of methods for drowsiness
and distraction detection are proposed and tested, it is very
difficult to compare them directly. This is because there are
no benchmark data sets available. To develop reliable
drowsiness detection systems, public data sets covering
simulated and real driving environments should be released
in the near future.
this problem, less intrusive sensors should be developed.

Currently, ECG signals can be captured using a less
intrusive manner. Driving performance behavior such as
steering wheel movement and standard deviation of lateral
position are also used to detect drowsiness.
Driver distraction is detected using head pose and gaze
direction. Driver distraction may lead to larger lane
variation, slower response to obstacles, and more abrupt
steering control. Thus, distraction should be monitored for
developing a safer driver-monitoring system.
For active driver safety systems, it is desirable to predict
unsafe driving behavior. We have explained prediction
methods based on facial expression and car dynamics.
Based on facial expression, the drivers emotion is detected,
which is helpful in predicting driving behavior.
Finally, we have discussed several issues to tackle in
future development of active driver safety systems. They
are a) hybrid measures for drowsiness detection, b) driving
context-awareness for safe driving, c) the availability of
public data sets for simulation and real driving conditions.
6. Conclusion
Acknowledgements
In this paper, we have reviewed the various methods

available to determine the drowsiness and distraction state
of the driver. Driver behavior such as visual features,
non-visual features and driving performance behavior are
explored to detect driver drowsiness. PERCLOS,
eye-closure duration (ECD), frequency of eye closure
(FEC) are visual feature-based systems used to detect
driver drowsiness. Among these, PERCLOS shows good
performance in detecting drowsiness but has some
limitations, such as illumination conditions. To overcome
this problem, an 850 nm IR illuminator is used.
Physiological signals such as ECG, EEG, EoG and PPG
signals are used as non-visual features to detect driver
drowsiness. Even though physiological signals show better
performance than visual features, they have some
limitations, particularly their intrusive nature. To overcome
Following are results of a study on the Leaders

Industry-university Cooperation Project, supported by the
Ministry of Education, Science & Technology (MEST).
Figure 4: Hybrid measurement for detecting driver state.
References
[1] U.S. Dept. of Transportation, Traffic Safety Facts 2006: A
Compilation of Motor Vehicle Crash Data from the Fatality
Analysis Reporting System and the General Estimates
System, tech. report DOTHS 810 818, Natl Highway
Traffic Safety Administration, 2006.
[2] http://www.lboro.ac.uk/departments/ssehs/research/behavio
ural-medicine/sleep-research-centre/
[3] Y. Liang, Detecting driver distraction, Ph. D thesis,
University of Iowa, 2009.
621
[4] M. Bayly, B. Fildes, M. Regan, and K. Young, Review of

crash effectiveness of intelligent transport system, TRaffic
Accident Causation in Europe (TRACE), 2007.
[5] E. Rogado, J. Garcia, R. Barea, L. Bergasa and E. Lopez,
Driver Fatigue Detection System, Proc. IEEE Int. Conf.
Robotics and Biomimetics, 2009.
[6] T. Nakagawa, T. Kawachi, S. Arimitsu, M. Kanno, K. Sasaki,
and H. Hosaka, Drowsiness detection using spectrum
analysis of eye movement and effective stimuli to keep
driver awake, DENSO Technical Review, vol. 12, pp.
113118, 2006.
[7] B. Hariri, S. Abtahi, S. Shirmohammadi, and L. Martel, A
Yawning Measurement method to Detect Driver
Drowsiness, Technical Papers, 2012
[8] C. Lin, L. Ko, I. Chung et al., Adaptive EEG-based
alertness estimation system by using ICA-based fuzzy neural
networks, IEEE Transactions on Circuits and Systems, vol.
53, no. 11, pp. 24692476, 2006.
[9] H. Cai and Y. Lin, An experiment to non-intrusively collect
physiological parameters towards driver state detection, in
Proceedings of the SAE World Congress, Detroit, MI, USA,
2007.
[10] Q. Ji, Z. Zhu, P. Lan, Real-Time Nonintrusive Monitoring
and Prediction of Driver Fatigue, IEEE Transactions on
Vehicular Technology, vol. 53, 2004.
[11] S. Abtahi, Driver Drowsiness Monitoring based on
Yawning Detection, MS thesis, University of Ottawa, 2012.
[12] F. Nasoz, O. Ozyer, C. Lisetti, and N. Finkelstein,
Multimodal affective driver interfaces for future cars, in
Proc. ACM Int. Multimedia Conf. Exhibition, pp. 319322,
2002.
[13] Y. Lin, H. Leng, G. Yang, H. Cai, An Intelligent
Noninvasive Sensor for Driver Pulse Wave Measurement.
Sensors Journal, IEEE, 2007.
[14] T. Victor, J. Harbluk, and J. Engstrm, Sensitivity of
eye-movement measures to in-vehicle task difficulty,
Transportation Research Part F, 8, 167-190, 2005.
[15] M. Jabon, J. Bailenson, E. Pontikakis, L. Takayama, and C.
Nass, Facial-Expression Analysis for Predicting Unsafe
Driving Behavior, Pervasive Computing, 2011.
[16] http://phys.org/news/2012-11-tracking-facial-features-safercomfortable.html#inlRlv
[17] B. Yin, X. Fan, Y. Sun, Multiscale dynamic features based
driver fatigue detection, Int. J. Pattern Recogn. Artif. Intell.
23, 575589, 2009.
[18] M. Akin, M. Kurt, N. Sezgin, M. Bayram, Estimating
vigilance level by using EEG and EMG signals, Neural
Comput. Appl., 2008.
[19] A. Kokonozi, E. Michail, I. Chouvarda, N. Maglaveras, A
Study of Heart Rate and Brain System Complexity and Their
Interaction in Sleep-Deprived Subjects, In Proceedings of
the Conference Computers in Cardiology, Bologna, Italy,
2008.
[20] Y. Guosheng, L. Yingzi, B. Prabir, A driver fatigue
recognition model based on information fusion and dynamic
Bayesian network, Inform. Sci., 2010.
[21] R. Khushaba, S. Kodagoda, S. Lal, G. Dissanayake, Driver
drowsiness classification using fuzzy wavelet-packet-based
feature-extraction algorithm, IEEE Trans. Biomed. Eng.,
2011.
[22] W. Liang, J. Yuan, D. Sun, M. Lin, Changes in

physiological parameters induced by indoor simulated
driving: Effect of lower body exercise at mid-term break,
Sensors, 2009.
[23] A. Sahyadehas, K. Sundaraj and M. Murugappan, Detecting
Driver Drowsiness based on Sensors: A Review, Sensors,
2012.
[24] K. Torkkola, N. Massey, and C. Wood, Driver inattention
detection through intelligent analysis of readily available
sensors, in Proc. Of IEEE Conf. on Intelligent
Transportation Systems, Washington, DC, pp. 326331,
2004.
[25] C. Liu, S. Hosking, M. Lenn, Predicting driver drowsiness
using vehicle measures: Recent insights and future
challenges, J. Saf. Res., 2009.
[26] L. Bergasa, J. Nuevo, M. Sotelo, R. Barea, M. Lopez,
Real-time system for monitoring driver vigilance, IEEE
Trans. Intell. Transport. Syst., 2006.
[27] T. DOrazio, M. Leo, C. Guaragnella, A. Distante, A visual
approach for driver inattention detection. Pattern Recog.
2007.
[28] W. Wierwille, L. Ellsworth, S. Wreggit, R. Fairbanks, C.
Kirn, Research on vehicle based driver status/performance
monitoring: development, validation, and refinement of
algorithms for detection of driver drowsiness, National
Highway Traffic Safety Administration Final Report, DOT
HS 808 247, 1994.
[29] D. Dinges, and R. Grace, PERCLOS: A valid
psychophysiological measure of alertness as assessed by
psychomotor vigilance, US Department of Transportation,
Federal highway Administration. Publication Number
FHWA-MCRT-98-006
[30] R. Grace et al., A drowsy driver detection system for heavy
vehicle, in Proc. 17th Digital Avionics Systems Conference,
Bellevue, USA, vol. 2, pp. I36/1-I36/8, 1998.
[31] C. Yan, Y. Wang, and Z. Zhang, Robust real-time
multi-used pupil detection and tracking under various
illumination and large-scale head motion, Computer Vision
and Image Understanding, 1223-1338, 2011.
[32] W. Shen, H. Sun, E. Cheng, Q. Zhu, Q. Li, Effective driver
fatigue monitoring through pupil detection and yawing
analysis in low light level environments. Int. J. Digit.
Technol. Appl., 2012.
[33] M. Flores, J. Armingol, A. de la Escalera, Driver
drowsiness warning system using visual information for both
diurnal and nocturnal illumination conditions, EURASIP J.
Adv. Signal Process., 2010.
[34] J. Jo, S. Lee, H. Jung, K. Park, and J. Kim, Vision-based
method for detecting driver drowsiness and distraction in
driver monitoring system, Optical Engineering, 2011.
[35] Seeing
machines
Driver
State
Sensor.
http://www.seeingmachines.com/product/ds
s/
[36] T. Xue, N. Nan, M. Fan, J. Yong, Head Pose Estimation
Using Isophote Features for Driver Assistance Systems, In
Proceedings of the IEEE Intelligent Vehicles Symposium,
Xian, China, 2009.
[37] E. Murphy-Chutorian, M. Trivedi, Head pose estimation
and augmented reality tracking: An integrated system and
evaluation for monitoring driver awareness, IEEE Trans.
Intell. Transp. Syst., 2010.
622
[38] T. Nakamura, T. Matsuda, A. Maejima, and S. Morishima,

Driver Drowsiness Estimation Using Facial Wrinkle
Feature, Siggraph poster, 2013.
[39] M. Miyaji, H. Kawanaka, K. Oguri, Drivers Cognitive
Distraction Detection Using Physiological Features by the
Adaboost, In Proceedings of the 12th International IEEE
Conference on Intelligent Transportation Systems, October,
2009.
[40] M. Patel, S. Lal, D. Kavanagh, P. Rossiter, Applying neural
network analysis on heart rate variability data to assess driver
fatigue, Exp. Syst. Appl., 2011.
[41] V. Vapnik, Support vector estimation of functions, in
Statistical Learning Theory, pp. 375570, Wiley, Hoboken,
NJ, 1998.
[42] S. Hu, G. Zheng, Driver drowsiness detection with eyelid
related parameters by support vector machine, Exp. Syst.
Appl., 2009.
[43] M. Kurt, N. Sezgin, M. Akin, G. Kirbas, M. Bayram, The
ANN-based computing of drowsy level. Exp. Syst. Appl.,
2009.
[44] X. Yu, Real-Time Nonintrusive Detection of Driver
Drowsiness, Technical Report for University of Minnesota:
Minneapolis, MN, USA, 2009.
[45] J. Hyun, S. Gih, K. Ko, S. Kwang, A Smart health
monitoring chair for nonintrusive measurement of biological
signals, IEEE Trans. Inform. Technol. Biomed., 2012.
[46] B. Lee, W. Chung, Multi-classifier for highly reliable driver
drowsiness detection in Android platform, Biomed. Eng.
Appl. Basis Commun., 2012.
[47] B. Cheng, W. Zhang, Y. Lin, R. Feng, X. Zhang, Driver
drowsiness detection based on multisource information,
Hum. Factors Ergon. Manuf. Serv. Indust., 2012.
[48] C. Wylie, T. Shultz, J. Miller, M. Mitler, R. Mackie,
Commercial Motor Vehicle Driver Fatigue and Alertness
Study: Technical Summary, FHWA-MC-97-001, 1996.
[49] H. Uno, Detection Decline in Arousal Level Using
Combined Physiological and Behavioral Measures, JARI
Research Journal Vol.25 No.8, 2003.
[50] S. Otmani, T. Pebayle, J. Roge, A. Muzet, Effect of driving
duration and partial sleep deprivation on subsequent
alertness and performance of car drivers, Physiol. Behav.,
2005.
[51] F. Ruijia, Z. Guangyuan, C. Bo, An on-Board System for
Detecting Driver Drowsiness Based on Multi-Sensor Data
Fusion Using Dempster-Shafer Theory, In Proceedings of
the International Conference on Networking, Sensing and
Control, Okayama, Japan, 2009.
[52] M. Ingre, T. Kerstedt, B. Peters, A. Anund, G. Kecklund,
Subjective sleepiness, simulated driving performance and
blink duration: Examining individual differences, J. Sleep
Res., 2006.
[53] A. Hattori, S. Tokoro, M. Miyashita, I. Tanakam, K. Ohue,
and S. Uozumi, Development of forward collision warning
system using the driver behavioral information, in
Proceedings of 2006 SAE World Congress, Detroit, MI,
USA, 2006.
[54] E. Murphy-Chutorian, A. Doshi, and M. Trivedi, Head pose
estimation for driver assistance systems: a robust algorithm
and experimental evaluation, in Proc. of 10th Int. IEEE
Conf. Intelligent Transportation Systems, pp. 709714,
2007.
[55] J. Kaminski, D. Knaan, and A. Shavit, Single image face

orientation and gaze detection, Mach. Vis. Appl. 21, 8598,
2009.
[56] E. Murphy-Chutorian and M. Trivedi, Head pose estimation
in computer vision: a survey, IEEE Trans. Pattern Anal.
Mach. Intell. 31(4), 607626, 2009.
[57] W. Hansen and Q. Ji, In the eye of the beholder: a survey of
models for eyes and gaze, IEEE Trans. Pattern Anal. Mach.
Intell. 32, 478500, 2010.
[58] S. Langton, H. Honeyman, and E. Tessler, The Influence of
Head contour and Nose Angle on the Perception of Eye-Gaze
direction, Perception and Psychophysics, vol. 66, no. 5, pp.
752-771, 2004.
[59] K. Ohue, Y. Yamada, S. Uozumi, S. Tokoro, A. Hattori, and
T. Hayashi, Development of a new pre-crash safety
system, presented at the Soc. Automotive Eng. World
Congr., Detroit, MI, 2006, SAE Technical Paper Series,
Paper 2006-01-1461.
[60] S. Lee, J. Jo, H. Jung, K. Park, and J. Kim, Real-time gaze
estimator based on drivers head orientation for forward
collision warning system, IEEE Trans. Intell. Transp. Syst.
12(1), 254267, 2011.
[61] M. Trivedi, T. Gandhi, and J. McCall, Looking-In and
Looking-Out of a Vehicle: Computer-Vision-Based
Enhanced Vehicle Safety, IEEE Trans. Intelligent
Transportation Systems, vol. 8, no. 1, 2007.
[62] A. Williamson and T. Chamberlain, Review of On-Road
Driver Fatigue Monitoring Devices, NSW Injury Risk
Management Research Centre, Univ. of New South Wales,
2005.
[63] A. Williamson and T. Chamberlain, Review of On-Road
Driver Fatigue Monitoring Devices, NSW Injury Risk
Management Research Centre, Univ. of New South Wales,
2005.
[64] L. Fletcher, L. Petersson, and A. Zelinsky, Road Scene
Monotony Detection in a Fatigue Management Driver
Assistance System, Proc. IEEE Intelligent Vehicles Symp.,
IEEE Press, 2005.
[65] J. McCall and M. Trivedi, Video-Based Lane Estimation
and Tracking for Driver Assistance: Survey, System, and
Evaluation, IEEE Trans. Intelligent Transportation Systems,
vol. 7, no. 1, 2006.
[66] M. Bertozzi et al., Knowledge-Based Intelligent
Information and Engineering Systems, B. Appolloi et

al., eds., Springer, 2007.
[67] P. Philip, P. Sagaspe, N. Moore, J. Taillard, A. Charles, C.
Guilleminault, B. Bioulac, Fatigue, sleep restriction and
driving performance, Accid. Anal. Prevent., 2005.
[68] J. Engstrm, E. Johansson, J. stlund, Effects of visual and
cognitive load in real and simulated motorway driving.
Transport. Res. Traffic Psychol. Behav., 2005.
623
Hindawi Publishing Corporation

International Journal of Vehicular Technology
Volume 2014, Article ID 437515, 15 pages
http://dx.doi.org/10.1155/2014/437515
Review Article
A Review of Active Yaw Control System for Vehicle Handling
and Stability Enhancement
M. K. Aripin,1 Yahaya Md Sam,2 Kumeresan A. Danapalasingam,2 Kemao Peng,3
N. Hamzah,4 and M. F. Ismail5
1
Control, Instrumentation & Automation Department, Faculty of Electrical Engineering, Universiti Teknikal Malaysia Melaka,
76100 Durian Tunggal, Melaka, Malaysia
2
Department of Control & Mechatronics, Faculty of Electrical Engineering, Universiti Teknologi Malaysia,
81310 UTM Johor Bahru, Johor, Malaysia
3
Temasek Laboratories, National University of Singapore 5A Engineering Drive 1, Singapore 117411
4
Faculty of Electrical Engineering, UiTM Pulau Pinang, 13500 Permatang Pauh, Pulau Pinang, Malaysia
5
Industrial Automation Section, Universiti Kuala Lumpur Malaysia France Institute, Section 14, Jalan Teras Jernang,
43650 Bandar Baru Bangi, Selangor, Malaysia
Correspondence should be addressed to M. K. Aripin; khairiaripin@utem.edu.my
Received 23 January 2014; Revised 8 May 2014; Accepted 15 May 2014; Published 12 June 2014
Academic Editor: Aboelmagd Noureldin
Copyright 2014 M. K. Aripin et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Yaw stability control system plays a significant role in vehicle lateral dynamics in order to improve the vehicle handling and stability
performances. However, not many researches have been focused on the transient performances improvement of vehicle yaw rate
and sideslip tracking control. This paper reviews the vital elements for control system design of an active yaw stability control
system; the vehicle dynamic models, control objectives, active chassis control, and control strategies with the focus on identifying
suitable criteria for improved transient performances. Each element is discussed and compared in terms of their underlying theory,
strengths, weaknesses, and applicability. Based on this, we conclude that the sliding mode control with nonlinear sliding surface
based on composite nonlinear feedback is a potential control strategy for improving the transient performances of yaw rate and
sideslip tracking control.
1. Introduction
In vehicle dynamic control of road-vehicle, controlling the
lateral dynamic motion is very important where it will
determine the stability of the vehicle. One of the prominent
approaches that are reported in the literature for lateral
dynamics control is a yaw stability control system. In order to
design an effective control system, it is essential to determine
an appropriate element of yaw stability control system. In
this paper, the elements of yaw stability control system, that
is, vehicle dynamic models, control objectives, active chassis
control, and its control strategies as depicted in Figure 1, are
extensively reviewed.
The linear and nonlinear vehicle models that described
the behaviour of lateral dynamic are explained for controller
design and evaluation purpose. To achieve the control objectives, it is essential to control the variables of yaw rate and
sideslip angle in order to ensure the vehicle stable. It is
required that the actual yaw rate and sideslip angle have
fast responses and good tracking capability in following
the desired responses. During critical driving condition or
manoeuvre, inappropriate commands by the driver to control
the steering and braking can cause the vehicle to become
unstable and lead to an accident. Therefore, an active control
for yaw stability control system is essential to assist the
driver to keep the vehicle stable on the desired path. By
implementing an active chassis control of steering or braking
or integration of both systems, the active yaw control system
can be realized.

Vehicle dynamics control
Vertical dynamics control
Longitudinal dynamics control
Lateral dynamics control
Yaw stability control system
Vehicle dynamics models
Lane warning detection system
Active chassis control
Control objectives
Lane keeping system
Control strategies
Figure 1: Yaw stability control system for vehicle lateral dynamic.
Vehicle dynamic models
Nonlinear model
7 DOF
8 DOF
Linearized model
14 DOF
2 DOF
Figure 2: Vehicle dynamic models.
In real driving condition, the lateral dynamics of vehicle

are incorporated with uncertainties such as different road
surface condition, varying vehicle parameters, and crosswind
disturbance. In yaw stability control system, these perturbations could influence the yaw rate and sideslip tracking
control performances. From the control system point of
view, the transient performances of tracking control are
essential. However, from the reviewed control strategies in
the literature, the controllers are not designed to cater this
matter. Therefore, an appropriate robust control strategy
should be proposed to improve the transient performances of
the yaw rate and sideslip tracking control in the presence of
uncertainties and disturbances. As a finding from the reviews,
this paper briefly discussed a possible high performance
robust tracking control strategy that can be implemented for
yaw stability control system.
The review begins with vehicle dynamics models in
Section 2. The yaw stability control objectives are discussed
in Section 3 and followed by active chassis control for in
Section 4. Yaw stability control strategies and problems are
reviewed in Sections 5 and 6, respectively. In Section 7, a high
performance robust tracking controller using sliding mode
control and composite nonlinear feedback is discussed. The
controller evaluation is discussed in Section 8 and ended with
conclusion in Section 9.
2. Vehicle Dynamics Models

In order to examine, analyse, and design the controller
for yaw stability control system, vehicle dynamics models
are essential where the mathematical modelling of vehicle
dynamic motion is obtained based on Newtons 2nd law
that describes the forces and moments acting on the vehicle
body and tires. In general, there are two categories of vehicle
dynamic model, that is, nonlinear vehicle model and linearized vehicle model as depicted in Figure 2. The following
subsections will discuss the nonlinear vehicle model for
simulation and linearized vehicle model for controller design
purpose.
2.1. Vehicle Model for Simulation. The nonlinear vehicle
model is regularly used to represent and simulate the actual
vehicle for controller evaluation and validation. In recent
years, researches in [15] have utilized nonlinear vehicle
model for vehicle handling and stability improvement studies. Figure 3 shows the typical nonlinear vehicle model in
cornering manoeuvre.
The input of this model is front wheel steer angle
while the output variables to be controlled are vehicle sideslip
and yaw rate . The vehicle parameters are vehicle width
track , distance from front, and rear axle to centre of

Lateral Motion. One has the following
d
Fx1 f
Fx2 f
Fy1
= (V + V )
= (1 + 2 ) sin + (1 + 2 ) cos + 3 + 4 .
(2)
Fy2
x
lf
= (1 cos + 2 cos + 1 sin + 2 sin )
r Mz
y
Yaw Motion. One has the following:
(3 + 4 ) + ,
Fx4
Fx3
(3)
lr
where is yaw moment that must be taken into account,

that is; > 0 if the tires tends to turn at -axis. In (2), the
lateral acceleration can be expressed in terms of vehicle
forward speed V, yaw rate , and sideslip as follows:
Figure 3: Nonlinear vehicle model [6].
= V + V = V ( + ) .
Fy3
Fy4
gravity (CG) and , respectively. The vehicle forward

velocity of centre of gravity (CG) is V, lateral velocity is
V , and longitudinal velocity is V . Other important vehicle
parameters are vehicle mass , moment of inertia , and
front/rear tire cornering stiffness / . The wheels are
numbered as subscript number with (1) for front-left, (2) for
front-right, (3) for rear-left, and (4) for rear-right.
Longitudinal tire force, , depends directly on tire slip
ratio, while lateral tire force, , depends directly on
tire sideslip angle, . For smaller slip angle and slip ratio,
lateral tire force is described as a linear function of the tire
cornering stiffness and tire sideslip angle while longitudinal
tire force is described as a linear function of the braking
stiffness and the tire slip ratio. For larger slip angle and slip
ratio, longitudinal and lateral tire forces exhibit a nonlinear
characteristics. Vehicle dynamic motion with nonlinear tire
forces represents a nonlinear system. The nonlinear lateral
and longitudinal tire forces can be described using prominent
Pacejka tire model as implemented in [1, 4, 7] or Dugoff tire
model as utilized in [810], while studies in [11] used both tire
models.
The nonlinear vehicle model could have different number
of degree-of-freedom (DOF) where it represents the dynamics motions and complexity of vehicle models. As utilized in
[2, 1214], the 7 DOF vehicle model represents the dynamic
motions of vehicle body, that is, longitudinal, lateral, yaw,
and four wheels. The dynamic equations for the longitudinal,
lateral, and yaw motions of the vehicle body are described as
follows.
Longitudinal Motion. One has the following:
= (V V )
= (1 + 2 ) cos + 3 + 4 (1 + 2 ) sin .
(1)
(4)
Therefore, the output variable of sideslip of two-track model

can be obtained as follows:
=
1
[cos (cos (1 + 2 ) sin (1 + 2 ))
V
sin (sin (1 + 2 ) sin (1 + 2 ))]
.
(5)
while the output variable of yaw rate can be determined

from (3) and obtained as follows:
=
1
[ ( cos + 2 cos + 1 sin + 2 sin )
1
(3 + 4 ) + ] .
(6)
In vehicle dynamic studies, each wheel represents 1 DOF.

Thus, there are 4 DOF for road-vehicle with 4 wheels. The
dynamic motion for each wheel is described as follows:
= + ,
(7)
where is wheel angular acceleration, , is wheel radius,

is wheel inertia, is braking torque, and is driving
torque.
Another nonlinear vehicle model used in the previous
research is 8 DOF vehicle model that is extensively used
in [4, 5, 911, 1518]. For more accurate simulation and
validation, the 14 DOF vehicle model is used in [1, 19, 20]. The
comparison between the number of DOF of nonlinear vehicle
models that have been discussed above can be summarized
and compared in Table 1.
Another nonlinear vehicle model used for simulation
uses a multi-degree-of-freedom vehicle model based on
commercial vehicle dynamics software, that is, CarSim, as

Table 1: Number DOF of nonlinear vehicle models.
Number of
Dynamic motions
DOF
7 DOF
(i) Longitudinal
(ii) Lateral
(iii) Vertical
(iv) Rotational of 4 wheels
8 DOF
(i) Longitudinal
(ii) Lateral
(iii) Vertical
(iv) Roll
(v) Rotational of 4 wheels
14 DOF
Fyr
Output variable
Fyf
f f
x
r
lf
lr
Yaw rate & Sideslip
Yaw rate, roll rate, and

sideslip
(i) Longitudinal
(ii) Lateral
(iii) Vertical
Yaw rate, roll rate, pitch
(iv) Roll
rate, and sideslip
(v) Pitch
(vi) Bounce
(vii) Rotational of 4 wheels
(viii) Vertical oscillations of
4 wheels
Figure 4: Bicycle model [31].
In the simplest form of planar motion, this model consists of 2

DOF for lateral and yaw motions as describe in the following
equations.
Lateral Motion. One has the following:
V ( + ) = ( + ) .
(8)
Yaw Motion. One has the following:

= .
(9)
In this model, the front and rear lateral tire forces and ,
respectively, exhibit linear characteristics and described as a
linear function of the front and rear cornering stiffness,
and , as follows:
= ,
implemented in [2126]. By using this software based vehicle

model, the dynamic behaviour of vehicle is more precise
similar to a real vehicle. However, for yaw rate and sideslip
tracking control in yaw stability control system, the 7 DOF
nonlinear vehicle model as discussed in the above equations
and shown in Table 1 is adequate for simulation and evaluation of the design controller.
2.2. Vehicle Model for Controller Design. In vehicle dynamic
studies, the classical bicycle model as shown in Figure 4
is prominently used for yaw stability control analysis and
controller design as utilized in [1, 3, 8, 2630]. This model
is linearized from the nonlinear vehicle model based on the
following assumptions.
where the front and rear tire sideslip angle, , and for
linear tire forces are given in the following equations:
=
(iii) Left and right wheels at the front and rear axle are
lumped in single wheel at the centre line of the vehicle.
(iv) Constant vehicle speed i.e. the longitudinal acceleration equal to zero ( =0)
(v) Steering angle and sideslip angle are assumed small
( 0).
(vi) No braking is applied at all wheels.
(vii) Centre of gravity (CG) is not shifted as vehicle mass
is changing.
(viii) 2 front wheels have the same steering angle.
(ix) Desired vehicle sideslip is assumed to be zero in
steady state.
,
(11)
= + .
V
By rearranging and simplifying (8)(11), the differential

equations of sideslip and yaw rate variables can be simplified
as a linear state space model as follows:
= + ,
(i) Tires forces operate in the linear region.

(ii) The vehicle moves on plane surface/flat road (planar
motion).
(10)
= ,
[ ] = [ 11 12 ] [ ] + [ 1 ] ,
21 22
2

1 +
]
[
V
V2
]
[
[ ]= [
][ ]
[
2 2 ]
[
(12)
[ V ]
]
[
+[
] ,
[ ]
[ ]
where and are state or output variables, and are front
and rear tire cornering stiffness, respectively, is vehicle
5
is essential to control both yaw rate and sideslip responses.
In order to achieve these control objectives, the proposed
controller must be able to perform the control task of the yaw
rate and sideslip tracking control.
Control objectives
Side slip
Yaw rate
Yaw rate and side slip
4. Active Chassis Control
Figure 5: Yaw stability control objectives.
mass, is moment of inertia, and are distance from

front and rear axle to centre of gravity, respectively, V is
vehicle speed, and front tire steer angle is the input
to the model. Notice that vehicle speed V is assumed always
constant which means the vehicle is not involved with accelerating and braking. Hence, only lateral and yaw motions are
analysed.
Besides that, the bicycle model is also regularly used as
desired or reference model to generate the desired response of
the yaw rate and sideslip angle based on steady state condition
or approximated first order response. In designing the control
strategy based on vehicle active chassis control, the linear
state space model in (13) is essential.
3. Yaw Stability Control Objectives

A vehicle yaw rate and sideslip angle are significant
variables in vehicle yaw stability control system. As stated in
[32], control objectives of yaw stability control system may
be classified into three categories, that is, yaw rate control,
sideslip control, and combination of yaw rate and sideslip
control as illustrated in Figure 5.
One of the control objectives of yaw stability control
system is yaw rate, r. An ability to control the actual yaw
rate close to desired response will improve the handling or
manoeuvrability of the vehicle. The desired yaw rate which
is generated by reference model should be tracked by the
controller in order to improve the handling performance as
mentioned in [2, 4, 13, 15, 18, 27, 33, 34]. In the steady state
condition, the desired yaw rate response can be obtained
by using the following equation:
=
V
( + ) + V2
(13)
where stability factor is depending on the vehicle parameters and defined as follows:
=
( )
( + )
(14)
Another control objective is the vehicle sideslip angle, ,

that is, the deviation angle between the vehicle longitudinal
axis and longitudinal axis and its motion direction. The
control of sideslip angle close to steady state condition means
controlling the lateral stability of the vehicle. For the steady
state condition, the desired sideslip is always zero, that is,
= 0 as mentioned in [1, 6, 9, 11, 17, 26, 35]. Therefore, to
improve the vehicle handling and stability performances, it
Steering and braking subsystems or actuator are part of the

vehicle chassis. The active control of yaw stability control
system can be realized through active chassis control, that
is, direct yaw moment control or active steering control or
integrated actives steering and direct yaw moment control
as shown in Figure 6. In direct yaw moment control which
can be implemented by active braking or active differential
torque distribution, the required yaw moment is generated by
the designed controller that controls the desired yaw rate and
sideslip. In active steering control, the wheel steer angle that
commanded by the driver is modified by adding corrective
steer angle from the designed controller. This control strategy
can be implemented either using active front steering (AFS)
or active rear steering (ARS) or four-wheel active steering
(4WAS) control. In order to control two variables of the yaw
rate and sideslip effectively, two different control mechanisms
are required. Thus, related research works on the integration
of two vehicle chassis control, that is, integrated active steering and direct yaw moment control, have been extensively
conducted recently. The review of direct yaw moment control,
active steering control, and integrated active steering and
direct yaw moment control are discussed in the following
subsections.
4.1. Direct Yaw Moment Control. Direct yaw moment control

is one of the prominent methods for yaw stability control
where extensive research works using this method have been
conducted with different control strategies and algorithms as
reported in [1, 3, 5, 8, 9, 1518, 25, 26, 30, 36]. It is recognized
as an effective method to enhance the vehicle lateral stability
during critical driving manoeuvre by controlling the slip ratio
of individual wheel. As illustrated in Figure 7, the required
corrective yaw moment, which is generated by the
transverse distribution of braking forces between the vehicle
wheels is calculated by the designed controller based on the
error between actual and desired vehicle model that have
been discussed in Section 2. Another approach of direct yaw
moment control is active distribution torque. By using an
active differential device as established in [19, 20, 37, 38],
the left-right driving torque is distributed by this device to
generate the required corrective yaw moment, .
As mentioned in Section 2, direct yaw moment control
design is based on the linear state space model. As described
in (15), is considered as control input and front steer angle
is assumed as disturbance:
[ ] = [ 11 12 ] [ ] + [ 1 ] + [ 3 ] ,
21 22
2
4
Direct yaw moment control
Active
braking
Integrated active steering and direct

yaw moment control
Active steering control
Active
differential
ARS
AFS
4WAS
Figure 6: Active chassis control.
fd
Desired
vehicle model
d , rd
Controller
Mz
, r
Actual vehicle
model
Figure 7: Direct yaw moment control [15].

1 +
]
[
V
V2
]
[
[ ]= [
][ ]
2
2
[
]
[
front wheel steers angle is a sum of steer angle commanded by

the driver and a corrective steer angle generated by the
controller. This corrective steer angle is computed based on
yaw rate and sideslip tracking errors 1 and 2 as implemented
in [6, 4347].
For control design and analysis of AFS control, the linear
state state space model as described in (16) is used. Noted that
this equation is similar to equation (12) but the front wheel
steer angle, = + :
0
[ V ]
]
[
+[
] + [ 1 ] .
[ ]
[ ]
[ ]
(15)
Although direct yaw moment control could enhance the
vehicle stability for critical driving conditions, it may be less
effective for emergency braking on split road surface. At
high vehicle speed steady state cornering, direct yaw moment
control could decrease the yaw rate and increase a burden
to the driver. To overcome this disadvantage, active steering
control is proposed.
4.2. Active Steering Control. Active steering control is another
approach to improving the vehicle yaw stability, especially
for steady state driving condition where the lateral tire
force is operated in the linear region. Research works of
active steering control have been continuously conducted in
order to improve the handling and stability performances as
reported in [7, 13, 3942]. In general, active steering control
can be divided into three categories, that is, active front
steering (AFS) control, active rear steering (ARS) control,
and four-wheel active steering (4WAS) control, as shown in
Figure 6. As road-vehicle normally has front-wheel steering,
AFS control becomes favourite approach among researchers
as it can be combined with active braking and/or suspension
control. In the AFS control diagram, as shown in Figure 8, the
[ ]=

1 +
]
[
V
V2
]
[
[
2
2 ] []
[
]
[
(16)
[ V ]
]
[
+[
] ( + ) .
[ ]
[ ]
On the other hand, ARS control is used to improve the
vehicle response for low speed cornering manoeuvres with
the input to the control system being the rear steering angle
. In order to enhance the manoeuvrability at low speed
and the handling stability at high speed, combination of AFS
control and ARS called 4WAS control has been proposed as
implemented in [24, 48, 49]. By implementing 4WAS control,
the lateral and yaw motion can be controlled simultaneously
using two independent control inputs. Noting that front
wheel steer angle and rear wheel steer angle with the
rear axles of rear tire cornering stiffness, and distance from
rear axle to centre of gravity are taken into account in the
input matric.
Desired
vehicle model
fd
e1 d
d , rd e = r r
2
d
fd + c
Controller
Actual vehicle
model
, r
Figure 8: Active front steering control [45].

fd + c
e1
Desired
vehicle model
fd
d , rd
e2
d
r rd
c
Controller
Mz
Actual vehicle
model
, r
Figure 9: Integrated active front steering-direct yaw moment control [53].
4.3. Integrated Active Chassis Control. The integrated active

chassis control has become a popular research topic in vehicle
dynamics control as discussed in [50]. Vehicle dynamics
control can be greatly achieved by integrating the active
chassis control of active steering, active braking, and active
suspension or active stabiliser as implemented in [12, 23, 51,
52]. Since road-vehicle is usually equipped with front-wheel
steering and braking system, an integration and coordination
of active front steering and direct yaw moment control are
the favourite approaches to achieving the objectives of yaw
rate and sideslip control as reported in [2, 10, 11, 27, 28, 53
59]. In this approach, the corrective front wheel steers angle
and corrective yaw moment are considered as two
independent control inputs to the vehicle as illustrated in
Figure 9.
For controller analysis and design of integrated active
front steering-direct yaw moment control, the linear state
space model used is describe as follows:

[ ]=
[
V
[
[
[
[
[ V
[
+[
[
[
1 +

V2
2 2
V
]
]
] []
]
]
(17)
0]
[ V ]
]
]
[
] [ ] + [
] .
[ ]
1]
]
[ ]
The principle of active chassis control of steering and braking

for yaw stability control has been discussed. From the
above discussion, the differences, advantages, and disadvantages of each active chassis control can be digested as
tabulated in Table 2. From this table, it can be observed
that, by implementing integrated active front steering-direct

yaw moment control, the lateral and yaw motions can be
controlled simultaneously using two independent control
inputs from two different actuators, that is, steering and
braking. Thus, this approach could enhance the vehicle yaw
stability where the yaw rate and sideslip can be controlled
effectively in emergency manoeuvres and the steady state
driving condition.
As a conclusion, active chassis control is essential for
active yaw stability control system. Therefore, to achieve the
yaw stability control objectives, the control strategies for yaw
rate and sideslip tracking control are developed based on this
active chassis control. The following section will review and
discuss the control strategies and algorithms that have been
developed in the past.
5. Yaw Stability Control Strategies

From the literature, various control strategies have been
explored and utilized based on particular algorithm for active
yaw stability control such as classical PID controller in [1],
LMI based and static state feedback control in [2, 8, 33],
control theory in [4, 13, 25], sliding mode control (SMC) in
[1, 7, 23, 24, 35, 38, 53], optimal guaranteed cost coordination
controller (OGCC) in [10], adaptive based control in [11],
mixed-sensitivity minimization control techniques in [16],
classical controllers PI in [49, 60], internal model control
(IMC) in [37], quantitative feedback theory (QFT) in [45],
and -synthesis control in [48]. Besides that, a combination
or integration of different two control schemes to ensure
the robustness of yaw stability control has been explored
such as SMC and backstepping method in [3], SMC and
Fuzzy Logic Control in [12], and LQR with SMC in [17]. As
discussed in [20], the IMC and SMC algorithms are designed
Steering and brake
Steering
Brakes
Vehicle actuator
4 wheels active steering

(4WAS) control
Active rear steering (ARS)

control
Integrated AFS-DYC control
Active steering control

(ASC)
(i) Effective for critical driving condition

(ii) Good for sideslip/wheelslip control
Advantages
Less effective during critical driving condition

Effective for critical and steady state driving condition
(i) Two different steer inputs

(ii) Goof for yaw rate control
(i) Two different inputs from two different
actuator (steering and braking)
(ii) Good for yaw rate and sideslip control
Disadvantages
(i) Less effective for braking on split road surface
(ii) Decrease yaw rate during steady state driving condition
(iii) Active differential need extra devices
(i) Rear wheel steer angle can be controlled

(ii) Good for yaw rate control
(i) Effective for steady state driving condition

Active front steering (AFS)
(ii) Ease to integrate with braking control
control
(iii) Good for yaw rate control
Direct yaw moment control Active braking

(DYC)
active differential
Table 2: Types of active chassis control.
8

for yaw stability control and the controllers performances are
compared and evaluated.
The control strategies are designed based on active chassis
control as discussed in Section 4. In active braking or active
differential which operates based on direct yaw moment
control (DYC), various robust control strategies have been
designed. As reported in [3], yaw stability control that
consists of tire force observer and cascade controller that is
based on sliding mode and backstepping control method is
designed. To solve the external disturbance as discussed in
[16], the robustness of mixed-sensitivity yaw stability controller is guaranteed for external crosswind and emergency
manoeuvres. To cater the uncertainty from longitudinal tire
force, the controller for wheel slip control is designed using
SMC algorithm for vehicle stability enhancement [17]. As
discussed in [20], the second order sliding mode (SOSM)
and enhanced internal mode control (IMC) are designed
as feedback controller to ensure the robustness against
uncertainties and control saturation issues. Both controllers
performances are compared and analysed for yaw control
improvement based on rear active differential device. Besides
that, the sliding mode control algorithm is also utilized to
determine the required yaw moment in order to minimize
the yaw rate error and side-slip angle for vehicle stability
improvement [22]. To overcome the uncertainties parameters
and guarantee robust yaw stability in [25], the control strategy
that consists of disturbance observer to estimate feedforward
yaw moment and optimal gain-scheduled is designed.
In the study of [30], the robust yaw moment controller and
velocity-dependent state feedback controller are matrixed by
solving finite numbers linear matric inequality (LMI). By
using this approach, the designed controller is able to improve
the vehicle handling and lateral stability in the presence of
uncertainty parameters such as vehicle mass, moment of
inertia, cornering stiffness, and variation of road surfaces and
also control saturation due to the physical limits of actuator
and tire forces.
In active steering control, robust control strategies are
designed to overcome the uncertainties and external disturbance problems. In [7], adaptive sliding mode control
is utilized to estimate the upper bounds of time-derived
hyperplane and uncertainties of lateral forces. As discussed
in [13], feedback control is implemented for robust
stabilization of yaw motion where speed and road adhesion
variations are considered as uncertainties and disturbance
input. As reported in [49], a proportional active front steering
control and proportional-integral active rear steering control
are designed for four-wheel steering (4WS) vehicle with
the objective to overcome the uncertainties of vehicle mass,
moment of inertia, and front and rear cornering stiffness
coefficients. To ensure a robust stability against system uncertainties, the automatic path-tracking controller of 4WS vehicle based on sliding mode control algorithm is designed [24].
In this study, the cornering stiffness, path radius fluctuation,
and crosswind disturbance are considered as uncertainty
parameters and external disturbance. As reported in [42], the
model reference adaptive nonlinear controllers is proposed
for active steering systems to solve the uncertainties and
nonlinearities of tires lateral forces. Quantitative feedback
9
theory (QFT) technique is implemented for robust active
front steering control in order to compensate for the yaw rate
response in presence of uncertainties parameters and reject
the disturbances [45]. As discussed in [48], robust controller
for 4WS vehicle is also designed based on -synthesis control
algorithm which considers the varying parameters induced
by the vehicle during driving conditions as uncertainties
while the study in [60] designed the steering control of vision
based autonomous vehicle based on the nested PID control
to ensure the robustness of the steering controller against the
speed variations and uncertainties of vehicle parameters.
In integrated active chassis control, an appropriate control
scheme is designed to meet the control objectives. Studies
in [2, 27, 33] have designed the control scheme that consists
of reference model based on linear parameter-varying (LPV)
formulation and static-state feedback controller with the
objective to ensure the robust performance for integrated
active front steering and active differential braking control.
In these studies, tire slip angle, longitudinal slips, and vehicle
forward speeds are represented as uncertainty parameters.
As reported in [4], integrated robust model matching chassis
controller that integrates active rear wheel steering control,
longitudinal force compensation, and active yaw moment
control is designed using controller based on linear
matrix inequalities (LMIs) for vehicle handling and lane
keeping performance improvement. In integrated active front
steering-direct yaw moment control, an optimal guaranteed
cost control (OGCC) technique is utilized in [10]. In this
study, tire cornering stiffness is treated as uncertainty during
variation of driving conditions. As discussed in [11], an
adaptive integrated control algorithm based on direct Lyapunov method is designed for integrated active front steering
and direct yaw moment control with cornering stiffness is
considered as a variation parameter to ensure the robustness
of designed controller. As reported in [23], sliding mode
controller is utilized for stabilising the forces and moments
in integrated control schemes that coordinated the steering,
braking, and stabiliser. In this study, the integrated control
structure is composed of a main loop controller and servo
loop controller that computes and distributes the stabilizing
forces/moments, respectively.
From the above discussion, these control strategies and
algorithms can be summarized and compared in terms of
their active chassis control, control objective, advantages,
and disadvantages as tabulated in Table 3. In conclusion,
an appropriate control strategy must be designed based
on particular algorithm. Robust control algorithms such as
, SMC, IMC, OGCC, QFT. are essential to solve the
uncertainties and disturbance problems that influenced the
yaw stability control performances. It is revealed that the
designed controllers in the above discussion are able to track
the desired yaw rate and vehicle sideslip response considering
external disturbances and system uncertainty.
6. Yaw Stability Control Problems

In the real environments, the dynamics of road-vehicle is
highly nonlinear and incorporated with uncertainties. Vehicle motion with nonlinear tire forces represents a nonlinear
Yaw rate
Yaw rate
Integrated AFS-DYC
Integrated AFS-DYC
DYC
4WAS
DYC
AFS
4WAS
OGCC
Adaptive integrated control

Mixed-sensitivity minimization
control
PI controller
IMC
QFT
synthesis control
SMC-LQR
SMC-FLC
Integrated steering, brake, and

suspension
DYC
Yaw rate and sideslip
DYC, active steering
SMC
SMC-backstepping
Integrated AFS-active differential

Integrated chassis control, active steering Yaw rate
LMI static state feedback
Yaw rate, sideslip, and roll

angle
Yaw rate
Yaw rate
sideslip
DYC
PID controller
Control objective
Control algorithms
Uncertainties are not consider
Disadvantages
Robust for uncertainty
Robust for uncertainties and nonlinearities
Robust for nonlinearities
Robust for uncertainties
Robust for uncertainties, reject disturbance
Robust for uncertainty
Robust for uncertainty, reject disturbance
Transient response
improvement is not consider
Uncertainties are not

considered
Transient response
Robust for uncertainties, reject disturbance Transient response

robust for uncertainties and reject
disturbance
Advantages
Anti-wind-up strategy to avoid high
overshoot and large settling time
robust for uncertainties
Table 3: Yaw stability control algorithms.
10
7. High Performance Robust

Tracking Controller
In this section, a principle of possible robust tracking control
strategy with high performance that can be implemented for
yaw rate and sideslip tracking control is discussed. Based
on the literature, a sliding mode control with the nonlinear
sliding surface can be proposed to improve the transient
response of the yaw rate and sideslip tracking control in
presence of uncertainties and disturbances.
7.1. Sliding Mode Control (SMC). Sliding mode control (SMC)
algorithm that had been developed in the two last decades is
recognized as an effective robust controller to cater for the
matched and mismatched uncertainties and disturbances for
linear and nonlinear system. It is also utilized as an observer
for estimation and identification purpose in engineering
system. Various applications using SMC are successfully
implemented as numerous research studies and reports have
been published. In vehicle and automotive studies, SMC is
one of the prominent control algorithms that is used as a
robust control strategy as implemented in [3, 17, 38, 53, 61
63].
Sliding mode control design consists of two important
steps, that is, designing a sliding surface and designing the
control law so that the system states are enforced to the sliding
surface. The design of sliding surface is very important as it
will determine the dynamics of the system being control. In
conventional SMC, a linear sliding surface has a disadvantage
in improving transient response performance of the system
1.4
1.2
Output response
system where the tire dynamic exhibit nonlinear characteristics, especially during critical driving conditions such as
a severe cornering manoeuvre. The main problems of yaw
rate and sideslip tracking control are uncertainties caused
from variations of dynamics parameters as discussed in the
previous section such as road surface adhesion coefficients
[8, 13, 33, 37, 45], tire cornering stiffness [2, 8, 1012, 20,
24, 30, 48, 49], vehicle mass [20, 30, 38, 45, 49], vehicle
speed [2, 13, 45], and moment of inertia [30, 49]. Besides
that, an external disturbance such as lateral crosswind may
influence the tracking control of desired yaw rate and
sideslip response as reported in [4, 6, 13, 24]. Therefore,
appropriate control strategies and algorithms are essential
to overcome these problems as discussed in the previous
section.
From the view of control system engineering, the
transient response performances of tracking control are
very important. However, the control strategies and algorithms discussed above are not accommodated for transient
response improvement of the yaw rate and sideslip tracking
control in presence of uncertainties and disturbances. The
designed controllers are only sufficient to track the desired
responses in the presence of such problems. Hence, an
appropriate control strategy that could improve the transient
performance of robust yaw rate and sideslip tracking control
should be designed for an active yaw control system which
can enhance the vehicle handling and stability performances.
11
1
0.8
0.6
0.4
0.2
0
10
12
14
16
18
20
Time (s)
Lightly damped system: fast rise-time and large overshoot.
Heavily damped system: sluggish response and small overshoot.
CNF control system: varying damping ratio,
fast and smooth response.
Figure 10: CNF control technique for transient performances

improvement [75].
due to constant closed loop damping ratio. Therefore, a

nonlinear sliding surface that changes a closed loop system
damping ratio to achieve high performance of transient
response and at the same time ensure the robustness has
been implemented in [6469]. In these studies, the nonlinear
sliding surface is designed based on the composite nonlinear
feedback (CNF) algorithm.
7.2. Nonlinear Sliding Surface Based CNF. The concept of
varying closed loop damping ratio, which could improve
transient response for uncertain system, is based on composite nonlinear feedback (CNF) control technique. This
technique that has been established in [7074] is developed
based on state feedback law. In practice, it is desired that
the control system to obtain fast response time with small
overshoot. But in fact, most of control scheme makes a tradeoff between these two transient performance parameters.
Hence, the CNF control technique keeps low damping ratio
during transient and varied to high damping ratio as the
output response closed to the set point as illustrated in
Figure 10.
In general, the design of the CNF control technique
consists of linear and nonlinear control law as describe as
follows:
= [Linear ] + [Nonlinear ] ,
= [ + ] + [ (, ) ( )] ,
(18)
where is feedback matrix, is a scalar, is input matrix,

> 0 is a solution of Lyapunov equation, and (, ) is
12
nonlinear function which is not unique and can be chosen

from the following equations:
2
(, ) = () ,
(, ) = || ,
(, ) =
(19)
((1(0 )/(0 )) 1 ) .
1
1
Based on tracking error, a nonlinear sliding surface adapted

from the CNF control law for an active yaw control system
can be defined as follows:
:= () = [1 ] [
1 ()
],
2 ()
(20)
where
1 := (, ) ,
(21)
where 1 () and 2 () could represent the yaw rate and sideslip

tracking error, respectively, is an input matrix of the system,
and is the identity matrix. Then, the nonlinear sliding
surface stability can be determined using Lyapunov stability
analysis and implement in the designed control law of SMC.
Based on the above discussion, the SMC with nonlinear
sliding surface based on CNF technique could achieve high
performance for uncertain systems. It could improve the
transient response performance in the presence of uncertainties and external disturbances. In addition, it is found that this
control strategy has not yet been examined for vehicle yaw
stability control system and should be further investigated.
Therefore, this control technique has initiated a motivation
to implement it for robust yaw rate and sideslip tracking
control in active yaw control systems. It is expected that this
approach could improve the vehicle handling and stability
performances.
8. Controller Evaluations
In order to evaluate the performance of designing controller,
simulations of emergency braking and driving manoeuvres
with the nonlinear vehicle model are usually carried out
according to ISO or SAE standards. The pure computer
simulations, cosimulation with other software or hardware in
the loop simulations (HILS), are the common approaches to
conducting the yaw stability test with or without driver model
for open loop or closed loop analysis, respectively.
One of the typical emergency braking manoeuvres for
vehicle yaw stability test is split- braking as reported in
[2, 37, 60]. In this test, the step input of brake torque is
applied to the vehicle in forward motion with constant speed
on split road surface adhesion coefficient, , where one side
of the wheels is on low and the other sides of the wheels
are on high or vice versa. This test is performed to test
the vehicle straight ahead driving stability. Critical driving
manoeuvres are also another efficient way to test the yaw
and lateral stability performances. A step steer manoeuvre
can be implemented to evaluate the steady state and transient
behavioural response of the vehicle as conducted in [16, 53,

55, 63]. Similarly, the constant speed J-turn manoeuvre is also
conducted for such purpose as reported in [5, 8, 9, 15, 30, 33,
45]. Another type of critical driving manoeuvre is lane change
manoeuvre as implemented in [3, 5, 10, 11, 15, 20, 21, 23, 26, 45,
46, 53, 55]. This manoeuvre can be conducted for open loop
single lane change or closed loop double lanes change with
driver model, lane change on different road conditions, lane
change on split- road, and lane change with braking effect.
With steering angle input is in sinusoidal form, the transient
handling behaviour can be evaluated and vehicle yaw and
lateral stability can be analysed.
Another test manoeuvres that can be implemented for
yaw stability control are steer reversal test for transient
performance evaluation [16, 19, 20], constant speed steering
pad to evaluate the steady state vehicle performance [19,
20], steering wheel frequency sweep for the bandwidth, and
resonance peak analysis [20] and also fishhook manoeuvre as
mentioned in [2, 25, 27]. In order to evaluate the yaw stability
control system performance in the presence of disturbance,
a crosswind disturbance as reported in [4, 6, 20, 24] is
considered as external disturbance that can influence the
lateral dynamic stability.
During critical driving manoeuvres, the actual response
of vehicles yaw rate and sideslip is obtained and analysed
in presence of uncertainties and external disturbances. By
performing the test manoeuvres as discussed above, it can
be concluded that the ability of the designed controller to
track the desired response should be validated. The responses
are usually compared to uncontrolled vehicles responses and
other controllers for their steady state and transient response
performances.
9. Conclusion
This paper has extensively reviewed the elements of yaw
stability control system. In designing yaw stability controller,
all these elements, that is, vehicle models, control objectives,
active chassis control, and control strategies, play an important role that contributes to the control system performances.
For controller design and evaluation, a 2 DOF linear and
7 DOF nonlinear vehicle models are essential. In order to
improve the handling and stability performances, the yaw rate
and sideslip tracking control are the main objectives that must
be achieved by the design controller. To realize an active yaw
stability control, an active chassis control of steering, braking
or integration of both chassis could be implemented with an
appropriate control strategies and algorithms.
In real driving condition, the uncertainties and external
disturbance may influenced the yaw rate and sideslip tracking
control performances. Hence, the robust control algorithm is
necessary. Based on this review, it has been concluded that
sliding mode control (SMC) is the best robust controller to
address these problems. From the view of control system,
transient performances are very important for tracking control. However, an existing SMC configuration does not have
capability to improve this transient performance. To address
this issue, a nonlinear sliding surface of SMC is designed

based on composite nonlinear feedback (CNF) algorithm.
This is because the CNF algorithm has been proven in
improving transient performances as discussed above. For
future works, this control strategy will be implemented for
yaw stability control system and the transient performances
of yaw rate and sideslip tracking control will be evaluated and
compared with classical SMC and other controllers.
Conflict of Interests
The authors declare that there is no conflict of interests
regarding the publication of this paper.
Acknowledgments
The authors would like to thank to Ministry of Education of
Malaysia, UTeM, and UTM for the supports of the studies.
References
[1] B. Lacroix, Z. Liu, and P. Seers, A comparison of two control
methods for vehicle stability control by direct yaw moment,
Applied Mechanics and Materials, vol. 120, pp. 203217, 2012.
[2] S. C. Baslamisli, I. E. Kose, and G. Anlas, Handling stability
improvement through robust active front steering and active
differential control, Vehicle System Dynamics, vol. 49, no. 5, pp.
657683, 2011.
[3] H. Zhou and Z. Liu, Vehicle yaw stability-control system design
based on sliding mode and backstepping control approach,
IEEE Transactions on Vehicular Technology, vol. 59, no. 7, pp.
36743678, 2010.
[4] J. Wu, Q. Wang, X. Wei, and H. Tang, Studies on improving
vehicle handling and lane keeping performance of closedloop driver-vehicle system with integrated chassis control,
Mathematics and Computers in Simulation, vol. 80, no. 12, pp.
22972308, 2010.
[5] G. Tekin and Y. S. Unlusoy, Design and simulation of an integrated active yaw control system for road vehicles, International
Journal of Vehicle Design, vol. 52, no. 14, pp. 519, 2010.
[6] H. Ohara and T. Murakami, A stability control by active angle
control of front-wheel in a vehicle system, IEEE Transactions
on Industrial Electronics, vol. 55, no. 3, pp. 12771285, 2008.
[7] Y. Ikeda, Active steering control of vehicle by sliding mode
controlswitching function design using SDRE, in Proceedings
of the IEEE International Conference on Control Applications
(CCA 10), pp. 16601665, Yokohama, Japan, September 2010.
[8] H. Du, N. Zhang, and F. Naghdy, Velocity-dependent robust
control for improving vehicle lateral dynamics, Transportation
Research C: Emerging Technologies, vol. 19, no. 3, pp. 454468,
2011.
[9] B. L. Boada, M. J. L. Boada, and V. Daz, Fuzzy-logic applied
to yaw moment control for vehicle stability, Vehicle System
Dynamics, vol. 43, no. 10, pp. 753770, 2005.
[10] X. Yang, Z. Wang, and W. Peng, Coordinated control of AFS
and DYC for vehicle handling and stability based on optimal
guaranteed cost theory, Vehicle System Dynamics, vol. 47, no. 1,
pp. 5779, 2009.
[11] N. Ding and S. Taheri, An adaptive integrated algorithm for
active front steering and direct yaw moment control based on
direct Lyapunov method, Vehicle System Dynamics, vol. 48, no.
10, pp. 11931213, 2010.
13
[12] S.-B. Lu, Y.-N. Li, S.-B. Choi, L. Zheng, and M.-S. Seong,
Integrated control on MR vehicle suspension system associated
with braking and steering control, Vehicle System Dynamics,
vol. 49, no. 1-2, pp. 361380, 2011.
[13] S. Mammar and D. Koenig, Vehicle handling improvement by
active steering, Vehicle System Dynamics, vol. 38, no. 3, pp. 211
242, 2002.
[14] C. Zhao, W. Xiang, and P. Richardson, Vehicle lateral control
and yaw stability control through differential braking, in Proceedings of the International Symposium on Industrial Electronics
(ISIE 06), pp. 384389, July 2006.
[15] M. Mirzaei, A new strategy for minimum usage of external yaw
moment in vehicle dynamic control system, Transportation
Research C: Emerging Technologies, vol. 18, no. 2, pp. 213224,
2010.
[16] V. Cerone, M. Milanese, and D. Regruto, Yaw stability control
design through a mixed-sensitivity approach, IEEE Transactions on Control Systems Technology, vol. 17, no. 5, pp. 10961104,
2009.
[17] S. Zheng, H. Tang, Z. Han, and Y. Zhang, Controller design
for vehicle stability enhancement, Control Engineering Practice,
vol. 14, no. 12, pp. 14131421, 2006.
[18] E. Esmailzadeh, A. Goodarzi, and G. R. Vossoughi, Optimal
yaw moment control law for improved vehicle handling,
Mechatronics, vol. 13, no. 7, pp. 659675, 2003.
[19] M. Canale and L. Fagiano, Comparing rear wheel steering
and rear active differential approaches to vehicle yaw control,
Vehicle System Dynamics, vol. 48, no. 5, pp. 529546, 2010.
[20] M. Canale, L. Fagiano, A. Ferrara, and C. Vecchio, Comparing
internal model control and sliding-mode approaches for vehicle
yaw control, IEEE Transactions on Intelligent Transportation
Systems, vol. 10, no. 1, pp. 3141, 2009.
[21] S. Moon, W. Cho, and K. Yi, Intelligent vehicle safety control
strategy in various driving situations, Vehicle System Dynamics,
vol. 48, no. 1, pp. 537554, 2010.
[22] S. Yim, W. Cho, J. Yoon, and K. Yi, Optimum distribution of
yaw moment for unified chassis control with limitations on the
active front steering angle, International Journal of Automotive
Technology, vol. 11, no. 5, pp. 665672, 2010.
[23] D. Li, S. Du, and F. Yu, Integrated vehicle chassis control based
on direct yaw moment, active steering and active stabiliser,
[24] T. Hiraoka, O. Nishihara, and H. Kumamoto, Automatic pathtracking controller of a four-wheel steering vehicle, Vehicle
System Dynamics, vol. 47, no. 10, pp. 12051227, 2009.
[25] S.-H. Yon, O.-S. Jo, S. Yoo, J.-O. Hahn, and K. I. Lee, Vehicle
lateral stability management using gain-scheduled robust control, Journal of Mechanical Science and Technology, vol. 20, no.
11, pp. 18981913, 2006.
[26] S. H. Tamaddoni, S. Taheri, and M. Ahmadian, Optimal
preview game theory approach to vehicle stability controller
design, Vehicle System Dynamics, vol. 49, no. 12, pp. 19671979,
2011.
[27] S. C. Baslamisli, I. E. Kose, and G. Anlas, Gain-scheduled
integrated active steering and differential control for vehicle
handling improvement, Vehicle System Dynamics, vol. 47, no.
1, pp. 99119, 2009.
[28] P. Falcone, H. Eric Tseng, F. Borrelli, J. Asgari, and D. Hrovat,
MPC-based yaw and lateral stabilisation via active front
steering and braking, Vehicle System Dynamics, vol. 46, no. 1,
pp. 611628, 2008.
14
[29] W. Cho, J. Yoon, J. Kim, J. Hur, and K. Yi, An investigation into
unified chassis control scheme for optimised vehicle stability
and manoeuvrability, Vehicle System Dynamics, vol. 46, no. 1,
pp. 87105, 2008.
[30] H. Du, N. Zhang, and G. Dong, Stabilizing vehicle lateral
dynamics with considerations of parameter uncertainties and
control saturation through robust yaw control, IEEE Transactions on Vehicular Technology, vol. 59, no. 5, pp. 25932597, 2010.
[31] Q. Li, G. Shi, J. Wei, and Y. Lin, Yaw stability control using the
fuzzy PID controller for active front steering, High Technology
Letters, vol. 16, no. 1, pp. 9498, 2010.
[32] W. J. Manning and D. A. Crolla, A review of yaw rate and
sideslip controllers for passenger vehicles, Transactions of the
Institute of Measurement and Control, vol. 29, no. 2, pp. 117135,
2007.
[33] S. C. Baslamisli, I. E. Kose, and G. Anlas, Design of active steering and intelligent braking systems for road vehicle handling
improvement: a robust control approach, in Proceedings of the
IEEE International Conference on Control Applications (CCA
06), pp. 909914, Munich 2006.
[34] P. Yih and J. C. Gerdes, Modification of vehicle handling
characteristics via steer-by-wire, IEEE Transactions on Control
Systems Technology, vol. 13, no. 6, pp. 965976, 2005.
[35] B. Kwak and Y. Park, Robust vehicle stability controller based
on multiple sliding mode control, in Proceedings of the SAE
World Congress, SAE 2001-01-10602001, 2001.
[36] P. Raksincharoensak, T. Mizushima, and M. Nagai, Direct yaw
moment control system based on driver behaviour recognition,
[37] M. Canale, L. Fagiano, M. Milanese, and P. Borodani, Robust
vehicle yaw control using an active differential and IMC
techniques, Control Engineering Practice, vol. 15, no. 8, pp. 923
941, 2007.
[38] M. Canale, L. Fagiano, A. Ferrara, and C. Vecchio, Vehicle
yaw control via second-order sliding-mode technique, IEEE
Transactions on Industrial Electronics, vol. 55, no. 11, pp. 3908
3916, 2008.
[39] P. Falcone, F. Borrelli, J. Asgari, H. E. Tseng, and D. Hrovat,
Predictive active steering control for autonomous vehicle
systems, IEEE Transactions on Control Systems Technology, vol.
15, no. 3, pp. 566580, 2007.
[40] P. Falcone, F. Borrelli, H. E. Tseng, J. Asgari, and D. Hrovat, Linear time-varying model predictive control and its application to
active steering systems: stability analysis and experimental validation, International Journal of Robust and Nonlinear Control,
vol. 18, no. 8, pp. 862875, 2008.
[41] F. Borrelli, P. Falcone, T. Keviczky, J. Asgari, and D. Hrovat,
MPC-based approach to active steering for autonomous
vehicle systems, International Journal of Vehicle Autonomous
Systems, vol. 3, no. 24, pp. 265291, 2005.
[42] Y. Kawaguchi, H. Eguchi, T. Fukao, and K. Osuka, Passivitybased adaptive nonlinear control for active steering, in Proceedings of the 16th IEEE International Conference on Control
Applications (CCA 07), pp. 214219, October 2007.
[43] S. Singh, Design of front wheel active steering for improved
vehicle handling and stability, in Proceedings of the SAE
Automotive Dynamics & Stability Conference, SAE 2000-01-1619,
2000.
[44] W. A. H. Oraby, S. M. El-Demerdash, A. M. Selim, A. Faizz, and
D. A. Crolla, Improvement of vehicle lateral dynamics by active
front steering control, in Proceedings of the SAE Automotive
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
[57]
[58]
[59]
Dynamics, Stability & Controls Conference and Exhibition, SAE

2004-01-2081, 2004.
J.-Y. Zhang, J.-W. Kim, K.-B. Lee, and Y.-B. Kim, Development
of an active front steering (AFS) system with QFT control,
International Journal of Automotive Technology, vol. 9, no. 6, pp.
695702, 2008.
B. Zheng and S. Anwar, Yaw stability control of a steer-by-wire
equipped vehicle via active front wheel steering, Mechatronics,
vol. 19, no. 6, pp. 799804, 2009.
Q. Li, G. Shi, and J. Wei, Yaw stability control using the
fuzzy PID controller for active front steering, High Technology
Letters, vol. 16, no. 1, pp. 9498, 2010.
G.-D. Yin, N. Chen, J.-X. Wang, and L.-Y. Wu, A study
on -synthesis control for four-wheel steering system to
enhance vehicle lateral stability, Journal of Dynamic Systems,
Measurement and Control, Transactions of the ASME, vol. 133,
no. 1, Article ID 011002, 2011.
R. Marino, S. Scalzi, and F. Cinili, Nonlinear PI front and rear
steering control in four wheel steering vehicles, Vehicle System
F. Yu, D.-F. Li, and D. A. Crolla, Integrated vehicle dynamics
control-state-of-the art review, in Proceedings of the IEEE
Vehicle Power and Propulsion Conference (VPPC 08), pp. 835
840, Harbin, China, September 2008.
L. Fei and D. Zhaoxiang, Integrated control of automotive four
wheel steering and active suspenion systems based on unifrom
model, in Proceedings of the 9th International Conference on
Electronic Measurement and Instruments (ICEMI 09), pp. 3551
3556, Beijing, China, August 2009.
S. Zhou, L. Guo, and S. Zhang, Vehicle yaw stability control
and its integration with roll stability control, in Proceedings of
the Chinese Control and Decision Conference (CCDC 08), pp.
36243629, July 2008.
A. Hu and F. He, Variable structure control for active front
steering and direct yaw moment, in Proceedings of the 2nd
International Conference on Artificial Intelligence, Management
Science and Electronic Commerce (AIMSEC 11), pp. 35873590,
Zhengzhou, China, August 2011.
A. Hu and B. Lv, Study on mixed robust control for integrated
active front steering and direct yaw moment, in Proceedings
of the IEEE International Conference on Mechatronics and
Automation (ICMA 10), pp. 2933, Xian, China, August 2010.
Z. He and X. Ji, Nonlinear robust control of integrated vehicle
dynamics, Vehicle System Dynamics, vol. 50, no. 2, pp. 247280,
2012.
C. Ahn, B. Kim, and M. Lee, Modeling and control of an antilock brake and steering system for cooperative control on splitmu surfaces, International Journal of Automotive Technology,
vol. 13, no. 4, pp. 571581, 2012.
C. Poussot-Vassal, O. Sename, L. Dugard, and S. M. Savaresi,
Vehicle dynamic stability improvements through gainscheduled steering and braking control, Vehicle System
J. Tjonnas and T. A. Johansen, Stabilization of automotive
vehicles using active steering and adaptive brake control allocation, IEEE Transactions on Control Systems Technology, vol.
18, no. 3, pp. 545558, 2010.
C. Rengaraj and D. Crolla, Integrated chassis control to
improve vehicle handling dynamics performance, in Proceedings of the SAE World Congress and Exhibition, SAE 2011-010958, April 2011.

[60] R. Marino, S. Scalzi, and M. Netto, Nested PID steering control
for lane keeping in autonomous vehicles, Control Engineering
Practice, vol. 19, no. 12, pp. 14591467, 2011.
[61] T. Shim, S. Chang, and S. Lee, Investigation of slidingsurface design on the performance of sliding mode controller
in antilock braking systems, IEEE Transactions on Vehicular
Technology, vol. 57, no. 2, pp. 747759, 2008.
[62] Y. M. Sam, J. H. S. Osman, and M. R. A. Ghani, A class of
proportional-integral sliding mode control with application to
active suspension system, Systems and Control Letters, vol. 51,
no. 3-4, pp. 217223, 2004.
[63] N. Hamzah, Y. M. Sam, H. Selamat, and M. K. Aripin, GAbased sliding mode controller for yaw stability improvement,
in Proceedings of the 9th Asian Control Conference (ASCC 13),
Istanbul, Turkey, 2013.
[64] D. Fulwani, B. Bandyopadhyay, and L. Fridman, Non-linear
sliding surface: towards high performance robust control, IET
Control Theory and Applications, vol. 6, no. 2, pp. 235242, 2012.
[65] B. Bandyopadhyay, F. Deepak, I. Postlethwaite, and M. C.
Turner, A nonlinear sliding surface to improve performance
of a discrete-time input-delay system, International Journal of
Control, vol. 83, no. 9, pp. 18951906, 2010.
[66] B. Bandyopadhyay and D. Fulwani, A robust tracking controller for uncertain MIMO plant using non-linear sliding
surface, in Proceedings of the IEEE International Conference on
Industrial Technology (ICIT 09), Churchill, Australia, February
2009.
[67] B. Bandyopadhyay and D. Fulwani, High-performance tracking controller for discrete plant using nonlinear sliding surface,
IEEE Transactions on Industrial Electronics, vol. 56, no. 9, pp.
36283637, 2009.
[68] S. Mondal and C. Mahanta, A fast converging robust controller
using adaptive second order sliding mode, ISA Transactions,
vol. 51, no. 6, pp. 713721, 2012.
[69] S. Mobayen, V. Johari Majd, and M. Sojoodi, An LMI-based
finite-time tracker design using nonlinear sliding surfaces,
in Proceedings of the 20th Iranian Conference on Electrical
Engineering (ICEE 12), pp. 810815, Tehran, Iran, May 2012.
[70] Y. He, B. M. Chen, and W. Lan, On improving transient performance in tracking control for a class of nonlinear discrete-time
systems with input saturation, IEEE Transactions on Automatic
Control, vol. 52, no. 7, pp. 13071313, 2007.
[71] G. Cheng, K. Peng, B. M. Chen, and T. H. Lee, Improving
transient performance in tracking general references using
composite nonlinear feedback control and its application to
high-speed XY-table positioning mechanism, IEEE Transactions on Industrial Electronics, vol. 54, no. 2, pp. 10391051, 2007.
[72] Y. He, B. M. Chen, and C. Wu, Composite nonlinear control
with state and measurement feedback for general multivariable
systems with input saturation, Systems and Control Letters, vol.
54, no. 5, pp. 455469, 2005.
[73] B. M. Chen, T. H. Lee, K. Peng, and V. Venkataramanan,
Composite nonlinear feedback control for linear systems with
input saturation: theory and an application, IEEE Transactions
on Automatic Control, vol. 48, no. 3, pp. 427439, 2003.
[74] Z. Lin, M. Pachter, and S. Ban, Toward improvement of
tracking performancenonlinear feedback for linear systems,
International Journal of Control, vol. 70, no. 1, pp. 111, 1998.
[75] G. Cheng, B. M. Chen, K. Peng, and T. H. Lee, A MATLAB
toolkit for composite nonlinear feedback controlimproving
transient response in tracking control, Journal of Control
Theory and Applications, vol. 8, no. 3, pp. 271279, 2010.
15
International Journal of
Rotating
Machinery
Engineering
Journal of

http://www.hindawi.com
Volume 2014
The Scientific
World Journal
Volume 2014
Distributed
Sensor Networks
Journal of
Sensors
Volume 2014

Volume 2014

Volume 2014
Journal of
Control Science
and Engineering
Advances in
Civil Engineering

Volume 2014
Volume 2014
Submit your manuscripts at

Journal of
Journal of
Electrical and Computer

Engineering
Robotics

Volume 2014
Volume 2014
VLSI Design
Advances in
OptoElectronics
Navigation and
Observation
Volume 2014


Chemical Engineering
Volume 2014
Volume 2014
Active and Passive

Electronic Components
Antennas and
Propagation
Aerospace
Engineering

Volume 2014

Volume 2014
Volume 2014
Modelling &
Simulation
in Engineering
Volume 2014

Volume 2014
Shock and Vibration

Volume 2014
Advances in
Acoustics and Vibration

Volume 2014
Available online at www.sciencedirect.com
ScienceDirect
Energy Procedia 82 (2015) 1017 1023
ATI 2015 - 70th Conference of the ATI Engineering Association
A review of experimental techniques for NVH analysis on a

commercial vehicle
Maria Antonietta Panzaa,b *
Department of Industrial Engineering, University of NaplesFederico II, Via Claudio 21, 80125 Naples, Italy
Istituto Motori CNR, Via Marconi 4, 80125 Naples, Italy
Abstract
Noise and vibration of automotive vehicles are an increasingly important issue in the automobile industry, for implications on both
environmental noise pollution and comfort perceived by driver and passengers. Since noise and vibration performances affect the
overall image of a vehicle, they are now considered important factors in the entire vehicle design process. In this regard, an accurate
experimental evaluation of vehicle noise and vibration levels in both stationary conditions and urban driving conditions are often
necessary for undertaking a refinement process of vehicle sound quality, satisfying the development targets. This paper provides a
review of the main experimental techniques adopted for the measurement and analysis of noise and vibration in a commercial
vehicle. In order to monitor sound generation in terms of both air-borne and structure-borne noise, the source, transfer path and
receiver have to be investigated by noise measurement. For this purpose, different techniques can be employed. Some of them make
use of instrumentation (i.e. condenser microphone, sound level meter, sound intensity probe, acoustic holography) which requires
particular measurement environments, as it ensures sufficient measuring accuracy only in free field or anechoic conditions. On the
contrary, the more innovative pressure-velocity sound intensity method can be used for in-situ tests, without the need to create
acoustically treated environments. At the same time, vibration measurements by using proper transducers allow to minimize the
structure-borne vibration transmitted from vehicle power plant to chassis and body structure. Experimental modal analysis is
another standard tool in vehicle NVH development for determining the dynamic characteristics of a system, and therefore for
reducing the risk of failure or excessively high structural vibration or sound pressure levels.
Hence, in present work, noise and vibration frequency spectra of experimentally analyzed case studies are given, to illustrate how
accurate experimental tests are fundamental for identifying the generation causes of typical noise and vibration problems.
2015 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
2015 The Authors. Published by Elsevier Ltd.
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Selection and/or peer-review under responsibility of ATI
Peer-review under responsibility of the Scientific Committee of ATI 2015
Keywords:experimental tests, frequency analysis, noise, sound quality, vehicle refinement, vibration
1. Introduction
There are many sources of noise in a vehicle: in the past the engine was considered the most important, therefore
the first NVH studies were applied to reduce noise and vibrations generated by the engine and powertrain. As over the
years a strongly reduced level of noise has been reached for these systems, other sources of noise such as road noise
have become very significant. Besides, the increase of the vehicle speed has strongly enhanced the importance of
aerodynamic noise [1], as well. The engine provides a very important contribution to the annoying noise perceived
inside a vehicle, representing a great source of internal vibration, too. The vibrations derive from the reciprocating and
* Corresponding author. Tel.: +39-081-7177130.

E-mail address:mariaantonietta.panza@unina.it.
1876-6102 2015 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of the Scientific Committee of ATI 2015
doi:10.1016/j.egypro.2015.11.861
1018
Maria Antonietta Panza / Energy Procedia 82 (2015) 1017 1023
rotational masses such as pistons, connecting rods and shafts. Other sources of vibration come from the gearbox, the
differential and the structural vibrating modes of the exhaust system .In addition, there are many acoustic sources in
the powertrain system inducing airborne noise, and among them the most important are the intake and exhaust
tailpipes. They act as very efficient acoustic sources and can be considered as monopole sources. The vehicle
suspension system plays a key role in vibration transmission, being located in the structure-borne transmission path
between the roadtyre interaction and the vehicle body. The tyres have a dual role in road-noise generation and
transmission. On the other hand squeaks and rattles represent indirect noise sources, as they are generated by the
dynamic displacement of the surfaces of the dashboard and internal trimmings, caused by the overall vibrations in a
car. Besides the above main phenomena, there are other secondary noise sources, such as brakes, electrical and
mechanical accessories, etc.
It is important to consider that internal vehicle noise does not depend only on the acoustic and vibration sources;
key roles are also played by the different transmission paths between the sources and the receivers (i.e. the drivers and
passengers ears). In a vehicle there are two different categories of transmission paths, related to completely different
mechanisms of energy transmission: structure-borne and air-borne paths. Commonly in a car, the structure-borne noise
transmission path dominates at low frequency (<200 Hz) while the air-borne noise transmission path dominates above
500 Hz [2]. In the mid-frequency range, both transmission paths have usually the same level of importance.
In order to implement an improvement process of NVH performance of a vehicle, the knowledge of the main noise
sources and the transmission paths represents a fundamental aspect.
This paper provides an overview of the main experimental techniques employed for the identification of noise
sources in a vehicle. The description of the techniques will be supported by proper plots referred to experimentally
analyzed case studies under actual urban driving conditions. The aim is to point out the role of experimental activity in
identifying possible noise and vibration generation problems, which could affect vehicle sound quality and overall
harmony [3].
Nomenclature
AI
Articulation Index
NVH
Noise Vibration Harshness
EFR
Engine Firing Rate
EMA
Experimental Modal Analysis
f0
Engine firing frequency
Multiples of engine rotating frequency
Engine rotational speed
Ncyl
Number of engine cylinders
ODS
Operational Deflection Shapes
OMA
Operational Modal Analysis
PVL
Power Velocity Level
TPA
Transfer Path Analysis
Particle velocity magnitude
uref
Reference effective particle velocity
Integer equal to 1 (for a two-stroke engine) or 2 (for a four-stroke engine)
2. Experimental techniques for NVH analysis in vehicles

Sound generation and transmission in a vehicle takes place in terms of small pressure waves that travel in air (airborne noise), and vibrations that travel in vehicle structures (structure-borne noise). While air-borne noise involves a
physical mechanism from which sound is generated and radiated (i.e. the hot displacement of fluid mass in the exhaust
tailpipe), structure-borne noise is caused from a vibrating source that induces the acoustic energy to travel through
solid structures and to be then released as air-borne noise (i.e. the engine structural vibrations). The main targets of
NVH experimental techniques are to determine the characteristics of these two types of noise in a vehicle, as well as
the transmission paths to driver and passengers. Only an accurate identification and characterization of noise and
vibration sources could allow to realize new refinements in design and to validate the solutions, in order to improve
vehicle interior sound comfort.
To date noise mapping techniques (sound intensity), acoustic holography and beam forming, Transfer Path
Analysis, modal analysis, order tracking are considered the most relevant experimental techniques to analyze and
identify NVH sources in a vehicle [4].Some of these techniques employ instrumentation which requires specifically
constructed rooms such as anechoic or reverberant chambers. In fact, only in these particular controlled sound fields,
the noisiness of a sound source can be easily related to the measured sound pressure ensuring sufficient accuracy. It is
the case, for instance, of a typical condenser microphone, that if used in rooms that are neither anechoic nor
reverberant could provide incorrect results, whether positioned too close to the sound source (at a distance less than the
wavelength of the lowest frequency emitted from the source, that is the near field) or too far away from the source,
where reflections from walls and other objects may considerably alter the measurement accuracy. Other measurement
devices that need to be employed in free field conditions are the sound level meter and the traditional pressure-pressure
sound intensity probe. However, thanks to the development of more innovative measurement techniques, sound
intensity in-situ tests are now possible also where anechoic conditions are not applicable, like an engine bay or car
interior. This technique relies on the use of the Microflown p-u intensity probe, the combination of a pressure
microphone and a particle velocity transducer which directly measure close to the source surface the scalar value
'sound pressure' and the vector value 'acoustic particle velocity' in one spot, giving in output the sound intensity
[5].Due to the near field benefits [6]using the direct measure of the directional particle velocity, p-u probes are not
highly affected by background noise or reflections, and measurements can be undertaken in real operational situation.
P-u probe represents an effective tool for vehicle noise sources identification issues. Typical measurements can be
performed, for instance, on the engine compartment adopting the Scan & Paint method [7]. The acoustic signals of the
sound field are acquired by manually scanning the emitting surface with the p-u probe while a camera is positioned
toward the surface to film the scanning. The sensor position is extracted by applying automatic color tracking to each
frame of the recorded video. It is then possible to directly visualize sound variations across the space in terms of sound
pressure, particle velocity or acoustic intensity.
As an example, in the following figures results obtained scanning the engine bay of a four-cylinder four-stroke
engine at a particular stationary operating condition (2000 rpm), are shown [8]. Fig. 1, in particular, reports acquired
data processed in terms of Particle Velocity Level (PVL), which is by definition equal to:
u
(1)
>dB@
PVL 20 log10
u ref
where u is the particle velocity magnitude and uref represents the reference effective particle velocity of 50 nano-meter
per second. The engine radiated noise spectrum always contains strong excitations associated with the cylinder firings
rate. The strongest tone in the spectrum is generally the Engine Firing Rate (EFR), defined as:
N
(2)
N cyl >Hz @
EFR
60 H
where N is the engine rotational speed expressed in revolutions per minute (rpm), is an integer equal to 1 or 2 for a
two-stroke or a four-stroke engine respectively, and Ncyl represents the number of cylinders. The engine firing
frequencies can be also expressed in terms of k-multiples of the rotating frequency, the so-called engine orders:
N
(3)
f o k >Hz @
60
According to formula (3) in a four-cylinder four-stroke engine the EFR is also called the 2ndengine order, as the
frequency is two times (k=2) the engines rotation one.
As expected, in Fig. 1 the first PVL peak which is the dominant excitation in the spectrum represents the 2nd engine
order, whereas the subsequent three peaks correspond to multiples of the engine combustion frequency at 2000 rpm.
1019
1020
Fig. 1. Frequency spectrum of the particle velocity signal at 2000 rpm
This analysis can be also useful to identify occurrences not directly related to engine firings. For instance, in Fig. 1
the not negligible PVL peak around 500 Hzis likely due to a resonance effect of the system at this particular
frequency.
As a final result, it is possible to visualize the sound field across the scanned space. In this regard, Fig. 2 shows the
acoustic color map of particle velocity at the frequency of 66.4 Hz (the 2nd engine order). It is immediate to note that
the main acoustic emissions come from the internal combustion engine, which radiates up to about 95 dB.
Fig. 2. Acoustic map of particle velocity at 66.4 Hz at 2000 rpm
This kind of sensor can be successfully used also for other applications in the field of mapping and analyzing
automotive noise and vibration. Panel Noise Contribution Analysis, a well-known methodology for an air-borne
Transfer Path Analysis (TPA) in car interior through which pressure contribution from the individual panels at a
reference point can be very accurately calculated [9][10], or Near Field Acoustic Camera for sound source localization
during transients even used in non-anechoic conditions [6], or In-Situ Absorption method for the measurement of the
acoustic properties of materials [11][12], are typical tools employed for increasing the acoustic comfort quality in a
car.
Order tracking analysis is another experimental technique which can provide useful information about vibrations
and acoustic signals generated in vehicle drive train components, related to engine rotational speed and in particular to
engine ignition frequency. In this respect, Fig. 3 shows an example of the spectrogram of exhaust noise signal acquired
by a proper microphone located in the proximity of the tailpipe end, during engine run-up from approximately 1000 to
3500 rpm [13].
Fig. 3. 3D acoustic intensity map of exhaust noise signal in engine run-up condition from approximately 1000 to 3500 rpm
Such analysis allows to visualize the magnitude peaks of the spectrum lined up with the orders spectrum lines, and
gives the possibility to identify the operating conditions and the frequency range of interest where noise emissions are
predominant. By observing Fig. 3, it can be noted that most of the noise content is produced in a low frequency range
up to approximately 350 Hz, with more pronounced peak values up to 100 Hz for engine rotational speeds above 2700
rpm. These latter could depend on a resonance effect related to exhaust natural frequencies to be better investigated.
Structural measurement techniques play a key role in the vehicle development process, as well. In particular,
powertrain vibration isolation design, if carefully carried out, allows to minimize the structure-borne vibration
transmitted from power plant to chassis and body structure. The most common acceleration transducer is the
piezoelectric accelerometer which consists of a quartz crystal with a mass bolted on top. Accelerometers are able to
detect very small vibrations and not be damaged by large vibrations in a wide frequency range from 0.2 Hz to 10 kHz.
The main methods used for structural dynamics analyses are Experimental Modal Analysis (EMA), Operational Modal
Analysis (OMA) and Operational Deflection Shapes (ODS).While the objective of traditional EMA is the construction
of a mathematical model of the dynamic characteristics of the object under investigation by measuring the frequency
response function between a force and a response transducers [14], OMA utilizes only response measurements of the
structure in operational condition to identify modal characteristics. On the other hand, ODS allows to determine the
real-world forced deflection during operation. With advances in performance of the test equipment and computer
capabilities, modal analysis has become a standard tool in vehicle NVH development. The knowledge of the relevant
dynamic characteristics helps engineers to avoid resonances at frequencies with high excitation, and therefore to
reduce the risk of damage or strong structural vibration or sound pressure levels.
3. Vehicle sound quality
As explained, the knowledge and characterization of the different noise sources play a key role in the sound
package design of a vehicle. However, although the level of interior noise reduction is very important for the acoustic
comfort of the occupants, it is not sufficient by itself. Another important factor that has to be taken into account is the
quality of the noise. In recent years there have been many sound quality studies and analyses in order to define the
criteria for a noise that is not just quiet but also pleasant. Some of the most commonly used sound quality and
psychoacoustic metrics for users perception in a vehicle are Articulation Index (AI), Loudness and Sharpness [2].
The AI is a measure of the intelligibility of voice signals, expressed as a percentage of speech units that are
understood by the listener when heard out of context. There are different algorithms for the calculation of the AI. The
method used in the automotive field is based on the 1/3-octave band levels in dB(A) between 200 Hz and 6300 Hz,
taking strongly into account the frequency bands where the typical speech frequencies are more important.
1021
1022
The Loudness level refers to the perceived intensity of the sound. It is based on the statistical response of the ear as
a function of frequency for a statistically significant number of people. The unit of Loudness is sone, which allows
to visualize the parameter on a linear scale. A sine tone of the frequency 1 kHz with a level of 40 dB has by definition
a Loudness of 1 sone. The Loudness scale is distinguished by the fact that a tone which is perceived to have double the
Loudness on the Loudness scale is designated by a doubled sone-value. For steady state sounds, standardized
calculation procedures have been defined by Zwicker and Stevens and accepted as ISO standards [15][16][17]. For
temporally varying sounds, Zwicker has also proposed an approach taking into account temporal effects [18]. The
latter evolved until its acceptance as DIN 45631/A1 [19].
The Sharpness is another important measure that takes into account the frequency content of the noise. It is defined
as the ratio between the high frequency noise level and the overall level; it is not related to sound intensity, but it is
high for metallic noise components that are generally considered annoying. A high Sharpness is hence correlated with
bad quality of the vehicle. The unit of Sharpness is acum. Sharpness delineates human sensation in a linear manner
as well. The reference sound of 1 acum is a narrow-band noise, one critical band wide, and at a center frequency of 1
kHz and having a level of 60 dB. The Sharpness calculation is based upon the specific Loudness distribution of the
sound. Therefore the calculation of the Loudness affects the results of the Sharpness analysis. The standard DIN 45692
standardizes the metric calculation.
An example of poor sound quality is represented by data reported in the following figures, [20]. In Fig. 4(a) the two
lines on the graph represent the Time Varying Loudness measured in a four-cylinder engine vehicle by two
conventional microphones located at the driver ears position, during engine run-up on a chassis dynamometer (red and
green indicate left and right ear respectively).As highlighted in the circled region, the overall level, measured in this
case in terms of Time Varying Loudness, exhibits strong deviations from the ideal trend (dashed blue line) at particular
engine speed ranges, giving the impression of a vehicle not characterized by a smooth and refined ride. In effect, a
vehicle signature that is loud but grows linearly with engine and vehicle speed will be more acceptable than a quieter
signature that shows large deviations from its mean trend under the same test conditions.
Fig. 4. Time Varying Loudness vs. RPM (a) and Sharpness vs. RPM (b) during vehicle acceleration
Also Sharpness plot in Fig. 4(b), which refers to the same test, indicates a low vehicle interior sound quality at a
particular operating condition (around 2675 rpm), as it reaches a substantially high level.
The measure of the mentioned psychoacoustic parameters helps to identify the best solutions to improve NVH
performance in terms of good interior sound quality rather than only reduced noisiness. Solutions can regard the use of
appropriate materials which ensure adequate acoustic insulation and absorption of the vehicle cabin, as well as
structural damping.
4. Conclusion
Several are the experimental techniques available nowadays for the characterization of vehicles noise and vibration.
In this paper a review of the main relevant experimental methods has been provided together with experimental data
representations of real analyzed case studies, with the aim to demonstrate how accurate measurements are very
important starting points for identifying the causes of noise generation and for undertaking a successive improvement
process of vehicles NVH performances.
References
[1] Yinzhi, H., Yang, Z., & Wang, Y. (2013). Wind Noise Testing at Shanghai Automotive Wind Tunnel Center. In Proceedings of the FISITA
2012 World Automotive Congress (pp. 571-578). Springer Berlin Heidelberg.
[2] Xu Wang, Vehicle noise and vibration refinement, Woodhead Publishing, 2010.
[3] G. Cerrato, Automotive Sound Quality Powertrain, Road and Wind Noise, Sound &Vibration, April 2009.
[4] G. Cerrato and P. Goodes, Practical approaches to solving noise and vibration problems, Sound & Vibration, April 2011.
[5] Microflown Technologies Ebook, Intensity.
[6] Microflown Technologies Ebook, Near Field Acoustics.
[7] Daniel Fernandez Comesana, Scan-Based Sound Visualisation Methods using Sound Pressure and Particle Velocity, Thesis for the degree of
Doctor of Philosophy, University of Southampton Institute of Sound and Vibration Research, July 2014.
[8] D. Siano, M. Viscardi, M. A. Panza, Experimental Acoustic Measurements in Far Field and Near Field Conditions: Characterization of a
Beauty Engine Cover, Recent Advances in Fluid Mechanics and Thermal Engineering, ISBN: 978-1-61804-268-2, pp.50-57, 2014.
[9] Bertolini, C., Horak, J., Mantovani, M., & Lo Sinno, G. (2011, July). Interior panel contribution based on pressure-velocity mapping and
acoustic transfer functions combined with the simulation of the sound package. In INTER-NOISE and NOISE-CON Congress and Conference
Proceedings (Vol. 2011, No. 6, pp. 2155-2172). Institute of Noise Control Engineering.
[10] Grosso, A., Fernandez Comesana, D. and De Bree, H., "Further Development of the PNCA: New Panel Noise Contribution ReferenceRelated (PNCAR)," SAE Int. J. Passeng. Cars - Mech. Syst. 5(2):2012, doi:10.4271/2012-01-1539.
[11] Li, M., van Keulen, W., Tijs, E., van de Ven, M., & Molenaar, A. (2015). Sound absorption measurement of road surface with in situ
technology. Applied Acoustics, 88, 12-21.
[12] Tijs, E., & Druyvesteyn, E. (2012). An intensity method for measuring absorption properties in situ. Acta Acustica united with Acustica,
98(2), 342-353.
[13] D. Siano, M. A. Panza, An Experimental Analysis of Vehicle Exhaust Noise Signals by Means of Order Tracking Approach, Recent
Advances in Systems, Proceedings of the 19th International Conference on Systems (part of CSCC '15), Zakynthos Island, Greece, July 16-20, 2015,
Recent Advances in Electrical Engineering Series, Vol.52, p.502, ISBN: 978-1-61804-321-4, ISSN: 1790-5117.
[14] D. Siano, M. Viscardi, P. Napolitano, M. A. Panza, Numerical and Experimental Acoustic Performance Investigations of a High-Speed
Train Composite Sandwich Panel, WSEAS TRANSACTIONS on APPLIED and THEORETICAL MECHANICS, E-ISSN: 2224-3429, Volume 9, pp.
290-300, 2014.
[15] International Organization for Standardization, Method for Calculating Loudness Level, ISO-532-1975 (E).
[16] E. Zwicker et al, Program for Calculating Loudness According to DIN45631 (ISO532B), Journal Acoustic Society Jpn (E), Vol. 12, Nr.1,
1991.
[17] S.J. Stevens, Procedure for Calculating Loudness : Mark VII, J. Acoust. Soc. Am., Vol. 33, Nr.11, pp.1577-1585, 1961.
[18] E. Zwicker, Procedure for Calculating Loudness of Temporally Variable Sounds, J. Acoust. Soc. Am., Vol. 62, Nr. 3, pp 675-681, 1977.
[19] German standard with amendment, Ref No. DIN 45631/A1.
[20] D. Siano, M.V. Prati, M. A. Costagliola, M.A. Panza, Evaluation of Noise Level Inside Cab of a Bi-Fuel Passenger Vehicle, WSEAS
Transactions on Applied and Theoretical Mechanics, ISSN / E-ISSN: 1991-8747 / 2224-3429, Volume 10, 2015, Art. #24, pp. 220-226.
Biography
M. A. Panza was born in Cosenza in 23/06/1986 Italy, and graduated in Mechanical Engineering at
the University of Calabria, Italy in 2012.She is currently a PhD Student in Industrial Engineering at
University of Naples Federico II, Italy, in collaboration with CNR - Istituto Motori, Italy. She is
dealing with acoustic and vibration in automotive field.
1023
Eur. Transp. Res. Rev. (2015) 7: 29

DOI 10.1007/s12544-015-0179-z
ORIGINAL PAPER
A review of fuzzy methods in automotive

engineering applications
Valentin Ivanov 1
Received: 26 October 2014 / Accepted: 18 August 2015 / Published online: 27 August 2015
# The Author(s) 2015. This article is published with open access at SpringerLink.com
Abstract
Purpose The paper addresses evolution of fuzzy systems for
core applications of automotive engineering.
Methods The presented study is based on the analysis of bibliography dedicated to fuzzy sets and fuzzy control for ground
vehicles. A special attention is given to fuzzy approaches used
in the following domains of automotive engineering: vehicle
dynamics control systems, driver and driving environment
identification, ride comfort control, and energy management
of electric vehicles.
Results The bibliographical analysis, supplemented with statistics of relevant research publications, has allowed to allocate the most important fuzzy application cases for each domain. In particular, it concerns driver identification, humanmachine interface, recognition of road conditions, and controllers of vehicle chassis and powertrain systems. It is found out
that fuzzy methods have the primary use most of all for tasks
requiring identification and forecasting procedures, especially
in conditions of limited informational space. Additional observation that can be also derived from the presented survey
points to reasonable integration of fuzzy technique with other
control engineering methods to improve the performance of
automotive control systems.
Conclusions In the aggregate the performed review indicates
that fuzzy computing can be considered as a versatile tool for
automotive engineering applications of different nature.
* Valentin Ivanov
valentin.ivanov@tu-ilmenau.de
1
Automotive Engineering Group, Ilmenau University of Technology,

Ehrenbergstr. 15, 98693 Ilmenau, Germany
Keywords Fuzzy systems . Road vehicles . Automotive

control systems . Vehicle dynamics control . Electric vehicles
1 Introduction
Fuzzy logic, among other computational intelligence
methods, attracts increased attention of engineers and researchers involved in the development of complex control
solutions for road vehicles and their subsystems. A possibility
to overcome various nonlinear models and to use intuitive
logical rules makes fuzzy control and fuzzy identification procedures useful tools by rapid system prototyping and on early
design stages. It is of particular relevance for intelligent vehicle functions and systems requiring the prediction of manoeuvres dynamics, identification of driving environment parameters and objects, and interaction with the driver. For the mentioned topics, the development of models of objects and relevant controllers of on-board systems can meet essential problems with feasibility, sufficient level of complexity and other
features demanded for real-time automotive applications.
These issues can be efficiently handled with the computational
intelligence methods, which are known as reasonable and
powerful tools in solving non-well-posed analytical problems.
Among different computational intelligence methods like neural networks, swarm intelligence computing et al. the fuzzy
logic can be of special relevance for intelligent automotive
systems. The reason is that the fuzzy logic has good applicability for identification tasks and pre-emptive control under
strong presence of data and model uncertainty.
The above-listed arguments are motivated the study
outlined in this paper. The main goal of the study is to give a
survey of existing fuzzy methods and systems that are finding
applications in the automotive area. Because the subject of the
study is characterized by a high degree of interdisciplinarity,
29 Page 2 of 10
the discussed review can be of particular interest for researchers working on topics of Ground Vehicle Engineering,
Control Engineering, and Computational Intelligence.
Specific objectives of the presented work can be formulated
as follows:
Define research topics that are most relevant to automotive fuzzy systems and tools;
Identify the main application cases for each defined topic;
Estimate development trends for each defined topic;
Analyse particular matters of automotive fuzzy systems
from viewpoint of (i) industrial applications and (ii) integration with other control techniques.
The results of the performed survey are given in next sections in accordance with the listed objectives. The article includes also extensive bibliography aimed at the representation
of various research schools around the world in context of the
discussed topics.
2 Review background
The history of fuzzy systems in vehicle-related areas begins
with several research works of founders of fuzzy logic dedicated to practical applications of fuzzy sets. An example is the
intersection controller that reduced junction-related delay of
vehicle driving proposed by Pappis and Mamdani in [1]. Later
Sugeno and Murakami in [2] and Sugeno and Nishida in [3]
have introduced the fuzzy realization of drivers logic for the
cases of an automated car parking system and handling the
desired vehicle trajectory. Then, from early 1990s, more and
more automotive systems with fuzzy logic are subjected both
to academic and industrial areas. This resulted in a considerable amount of publications and patents.
An analysis of the published studies and works allows to
allocate a number of domains, where fuzzy logic has found the
most acceptance. These domains can be conditionally called
as BDriver^, BVehicle Dynamics^, BRide comfort^, BElectric
vehicles^, and BDriving environment^. Results of analysis of
publications relevant to the fuzzy modeling and control of
automotive systems in the listed engineering areas are further
introduced in the paper.
The presented review uses the methodology proposed and
partially implemented in [4]. The analysis uses only Englishlanguage scientific publications found in Scopus abstract and
citation database of peer-reviewed literature (http://www.
elsevier.com/online-tools/scopus). The following limitations
of the analysis are accepted within the framework of this
paper: The works are analyzed in relation to the ground and
road vehicles only. Mobile robots and machines with nonholonomic constraints are excluded from the analysis due to
their less relevance to automotive applications.
The overview is accompanied by statistics of peerreviewed journal articles from two recent decades (dated
19942013) in each domain. (Conference papers are excluded
from statistical analysis to avoid the consideration of duplicated content. The elimination has been also done for those papers, which content was already published in another journal.
This is because it was observed that in particular cases the
same content is repeated with minor variations in several conference papers published by the same authors.)
Additional section of the paper will analyse the publications of SAE International (http://www.sae.org) as one of the
most recognized informational source worldwide for
automotive engineering and presenting mainly the studies,
which are providing experimental verification and close to
industrial application.
3 Fuzzy methods for driver modelling and driver

assistance systems
The domain BDriver^, proposed for the classification, implicates topics related to the driver models, driver behavior, and
driver assistance systems. These topics are of special relevance for fuzzy applications because almost emotional and
psychological facets of human behavior carry more semantic
as numerical uncertainty. It complicates the formulation of the
driver in simulation and control tasks through non-soft computing methods. The first applications of fuzzy sets to the
modeling of the driver are arisen in 1980s. In particular,
Kramer and Rohr in [5] and Kramer in [6] proposed to use a
fuzzy model for the representation of perception characteristics of the driver and illustrated this approach on the driving
simulator. Nagai, Kojima and Sato [7] suggested the fuzzy
driver model describing the subjective recognition, judgment
and control of the vehicle speed. This model was built from
the accident analysis data. Then Kageyama and Pacejka [8]
and Ehara and Suzuki [9] developed the first fuzzy models of
the human reasoning for more complex driving situations.
Analysis of relevant journal publications, Fig. 1, and conference papers allows to allocate the most typical applications of
fuzzy methods within the domain DriverB:
Identifications and classifications of the drivers regarding

fatigue, emotions and other human attributes, including
the procedures of driver state recognition and forecasting
through monitoring of various physiological parameters
like electroencephalography-estimated brain activity, eye
movement, gestures et al. [1015],
Structure and controllers of driver assistance systems and
devices of human machine interface [1618],
Models of driver actions on vehicle controlling devices
(brake and throttle pedals, steering wheel) for authentic
Page 3 of 10 29
Fig. 1 Number of journal publications related to fuzzy methods in

domain BDriver^ and cited in Scopus database

domain BVehicle Dynamics^ and cited in Scopus database
simulation of vehicle maneuvers on driving simulators;

controllers of pedal and steering wheel robots [1922],
Simulation of driver reasoning for controllers of (semi-)
automated and unmanned ground vehicles [2325],
Understanding of subjective evaluation of vehicle dynamics; driver feeling of vehicle dynamics parameters
like velocity, road friction and other [2629],
Advisory functions of human-machine interface systems
supporting the driver in eco- and low-emission vehicle
operation [3032].
various numerical methods. Several early examples of fuzzy

computing for tire parameters estimation are described in works
of Stumpf, Arendt and Lux [33], and Madau, Yuan, Davis Jr. and
Feldkamp [34]. Among vehicle dynamics control systems, ABS
belongs to the classical examples of automotive control applications to verify and to define functional properties of different
control methods, including fuzzy computing. In particular, Intel
Corporation has proposed MCS 96 microcontroller family for
the first fuzzy braking processors [35]. Later Siemens AG has
applied fuzzy coprocessors C99A for brake-by-wire systems
[36]. Progress in vehicle dynamics control (VDC) systems has
given many opportunities for practical implementation of fuzzy
logic. Whereas the first VDC systems have used only brakes and
engine as actuators, the actual trend is to enable other chassis
systems such as steering and suspension in an integrated control
circuit. As a result, the domain BVehicle Dynamics^ includes a
series of the research problems where fuzzy methods found use:
From practical viewpoint, the fuzzy logic gives valuable

opportunity to develop controllers of human machine interface and driver assistance, which can be integrated with other
automotive systems responsible for active safety control
through the correction of vehicle dynamics by active chassis
and powertrain subsystems. The controllers of these subsystems can also implement different fuzzy methods that will be
discussed in next section.
Brake control including architecture and algorithms of

anti-lock braking systems [3740]
Vehicle traction control including solutions for engine
control, electronic differentials, powertrain and driveline
control in general [4143],
Control of lateral vehicle dynamics, in particular, in terms
of yaw rate and vehicle sideslip control [4448],
4 Fuzzy methods for vehicle dynamics control

and ride comfort
Within the framework of the presented review the domain

BVehicle Dynamics^ relates to the systems controlling the lateral
and longitudinal vehicle motion. The corresponding representatives are anti-lock braking (ABS) and traction control systems,
torque vectoring, electronic stability control, electronic differentials et al. The systems responsible for the vertical dynamics of the
vehicle are allocated to another domain BRide Comfort^ due to
considerable amount of relevant publications. Statistics of journal
papers for both mentioned domains is depicted in Figs. 2 and 3.
The first applications of fuzzy methods to the vehicle dynamics were related to the formalization of tire parameters, which are
used in the brake and acceleration control. It can be explained
with the fact that automotive tires have nonlinear characteristics
of friction and slip, which cannot be measured by on-board sensors and require therefore real-time estimation with the help of

domain BRide Comfort^ and cited in Scopus database
29 Page 4 of 10
Steering control including solutions for electric power

steering; active front and rear steering [4951]
Estimation of vehicle state (linear velocity, yaw rate, vehicle sideslip angle et al.) from the sensors and experimental data [5254],
Integrated sequential or parallel control on vehicle dynamics through independent subsystems (e.g. brakes,
steering, suspension, driveline) [5558],
Identification and estimation of parameters of tire-road
interaction [5963].
It was found during the analysis of research literature that

studies of semi-active and active suspension control take up a
larger share comparing with publications in other of automotive
systems based on fuzzy methods. The reason can be that the
ride control systems are characterized by complex nonlinear
oscillating behaviour influencing simultaneously several properties like comfort, handling, or NVH (noise, vibration, harshness). The use of soft computing methods can certainly simplify
the control on these vehicle properties. The earliest variants of
fuzzy architecture of active suspension controllers were proposed by Yester and McFall [64], and Lin and Lu [65]. Then
many other studies related to the fuzzy applications in the domain BRide comfort^ are arisen with the trend to continuous
growth of research in this area. The main topics within the
domain BRide comfort^ can be subjected as follows:
Methods of objective evaluation and control of the vehicle ride comfort [6668],
Control algorithms of suspension actuators and shock absorbers with adaptation to environmental and operational
conditions such as road roughness, vehicle dynamic variables and other [6973]
Special control strategies of electrorheological,
magnetorheological, and electrical dampers and specific
suspension elements [7477].
It should be noted that in spite of numerous publications in

the domain BRide comfort^, there are few reports about industrial variants of fuzzy suspension controllers. Overwhelming
majority of analyzed investigations belongs to the fundamental or conceptual applied research.
Now many classes of ground vehicles require the mandatory installation of VDC systems. This fact can encourage
more intensive investigations on fuzzy approaches in vehicle
dynamics and ride control.
information and communication technologies allowing closer

interaction of the vehicle with the driving environment.
Impactful control solutions improving vehicle safety and energy efficiency are being proposed now with the help of various on-board and on-road sensors as well road infrastructure
services. All these aspects including the technologies for
(semi-)autonomous driving are subjected to the domain
BDriving Environment^, which has been also analysed in the
presented study, Fig. 4. It should be mentioned that specific
topics of traffic control and architecture of Intelligent
Transport Systems are excluded from the review because they
are related mainly to Transport Engineering, but not to
Automotive Engineering.
The first research works relevant to fuzzy systems in the
domain BDriving Environment^ have investigated path planning algorithms for an autonomous vehicle supported by a
navigation system or equipped with advanced set of sensors,
which are able to percept external parameters like distance
between the vehicles or to identify external objects. The studies of Hogle and Bonissone [78] and Freisleben and
Kunkelmann [79] can be mentioned in this context. These
and other topics are in scope of standing research interest from
middle of 1990s and now the main subjects of fuzzy applications for the domain BDriving Environment^ can be classified
as follows:
Adaptive cruise control for the vehicles controlled by the

driver; road following control for automated driving; collision prevention systems [8082],
Processing sensor information; recognition of driving environment parameters; vehicle localization [8389],
Coordination of road vehicle platoon systems [9092],
Control architecture of autonomous vehicles [9396];
Specific problems of brake, traction and steering control
systems of autonomous vehicles [9799];
Automated parking systems [100102].
In accordance with the short- and long-term projection, the

mass-produced cars with functions of autonomous driving
5 Fuzzy methods and vehicleenvironment

interaction
Both driver assistance and vehicle dynamics control systems,
discussed in previous sections, can benefit from new

domain BDriving Environment^ and cited in Scopus database
will be stepwise introduced on the market in coming decade.

This fact should stimulate more intensive research in various
topics of the domain BDriving Environment^.
6 Fuzzy methods for electric vehicles

Hybrid electric vehicles, full electric vehicles, fuel cell
electric vehicles are at the centre of permanent attention
of researchers and developers of fuzzy systems. This
tendency is growing now because of sweeping development of technologies for BGreen Mobility^, Fig. 5. Early
studies have introduced fuzzy energy management (EM)
of electric vehicles in general. In particular, Farrall and
Jones in [103] and Cerruto, Consoli, Raciti and Testa in
[104] have proposed different fuzzy EM systems for
hybrid vehicle, which were responsible for efficient
powertrain control. Then, comparing with other automotive applications, fuzzy methods found many-sided use
in the domain BElectric vehicles^. An analysis of relevant research works allows to mention the following
applicative areas for fuzzy methods:
Global energy management of electric vehicles; hybrid

powertrain control of operational modes (internal combustion engine / electric motor) [105110];
Traction control, anti-lock braking and regenerative braking control of electric vehicles [111113];
Internal controllers of electric motors, starters, inverters;
electric propulsion controllers [114116];
Forecast and optimization of driving range [117, 118];
Estimation of battery performance and algorithms for battery charging [119121].
A number of other aspects like interaction of electric

vehicles with the road and urban infrastructure are
among further promising fuzzy applications in this research area.

domain BElectric Vehicles^ and cited in Scopus database
Page 5 of 10 29
7 Fuzzy methods for miscellaneous applications

In addition to previously introduced domains, various unclassified or rarely studied examples of automotive fuzzy systems
are also presented in analyzed research publications.
In particular, the most interesting variants cover the following topics: Transmission control including gear shifting algorithms and clutch control [122], Systems for diagnosing different vehicle systems and elements [123125], Methods for
the assessment of passive safety [126, 127], Thermal and
NVH comfort control systems [128, 129], Vehicle body design [130], etc. Figure 6 shows the statistics of relevant journal
publications. Hence, fuzzy methods can be considered as
well-established research tools in different aspects of automotive engineering.
8 Fuzzy methods in applied and industrial research

One of the interesting results of the analysis is that a
serious gap still persists between pure research and industrial applications of fuzzy methods in automotive engineering. To support this statement, Fig. 7 displays
statistics of publications related to industrial or readyto-installation specimens of road vehicle systems with
fuzzy controllers or fuzzy models embedded into the
processing units. This selection reflects the analysis of
technical papers published by SAE International as one
of the most recognized information sources worldwide
for automotive engineering. The preliminary search has
found only few publications every year with information
about application of fuzzy methods in automotive area.
The deep analysis detected neither the clear-cut growth
of fuzzy applications nor their regularity in individual
domains. The reason for the available gap between theoretical studies and industrial systems can be explained
by the fact that most of traditional automotive control
systems have well-established variants of system architecture and corresponding re al-time func tional
Fig. 6 Number of journal publications related to fuzzy methods for

miscellaneous automotive applications and cited in Scopus database
29 Page 6 of 10
Fig. 7 Selected SAE

publications related to fuzzy
methods
realizations, which are most often based on rule-based,

nonlinear, optimal and other control approaches that are
not directly related to computational intelligence
methods. Therefore, until recently the progress in the
vehicle control systems has been related first of all to
the implementation of more powerful and efficient hardware accompanied with cautious modification of software part. However an increased demand for on-board
systems, communicating with the driver and performing
adaptation and learning in accordance with human behaviour, shows certain limitations in traditional, non-soft
computing control techniques.
Nevertheless, it should be noted that some applicative areas
for fuzzy methods are especially most commonly addressed in
published industrial research studies. This is particularly true
for on-board systems like the cruise control, lane change assistant et al., which require identification of the driver behaviour. The research of Ford Motor on the driver characterization
for the car following [131, 132] and for the vehicle destination
prediction [133] can be mentioned in this regard. A reasonable
applicability of fuzzy methods to this topic has been also demonstrated in studies with the participation of Centro Richerce
Fiat for the driver distraction modelling [134]. Other work
[135], performed under coordination of researchers from
Renault, has indicated an efficient implementation of fuzzy
sets and fuzzy space windowing for the psychophysiological characterization of drivers within the context
of the car following.
An analysis of industrial white papers and analytical publications indicates trends to further consistent growth of intelligent automotive systems with the logic based on fuzzy
methods [136]. This statement can be also confirmed with
numerous industrial patents for fuzzy algorithms and controllers, claimed by automotive OEMs and suppliers. It allows to
expect that the mentioned gap between pure research and industrial applications should be overcome in near future.
9 About compatibility of fuzzy methods

in automotive engineering
Eventual barriers for more intensive dissemination of fuzzy
methods in automotive engineering can be removed through
their integration with other analytical and numerical methods.
In particular, such integration can improve robustness and
adaptivity of controllers with simultaneous keeping of
straightforward formalization of control tasks. The fact that
fuzzy logic remains one of the most inviting approaches in
basic and applied sciences it can be supported by the important indicator of high compatibility of fuzzy methods with
other computing and control techniques. The analyzed research publications offer many examples of integration of
fuzzy sets and fuzzy logic that are summarized in Table 1.
10 Conclusions
Analysis of research literature, especially results of published
experimental works, demonstrates that fuzzy methods have
solid background and good prospects for the implementation
Table 1
Examples of methods used in automotive applications
Conventional fuzzy methods
Methods / tools integrated

with fuzzy technique
Takagi-Sugeno fuzzy model [137]

Fuzzy linear regression [9]
Fuzzy preview control [138]
Model reference adaptive fuzzy
control [44]
Neuro-fuzzy method [123]
Hierarchical fuzzy integral [24]
Fuzzy clustering [139]
Type-II fuzzy system [140]
Sliding mode control [141]

Grey predictor [142]
PI and PID control [91, 143]
Genetic algorithm [71, 144]
Optimization gradient method [145]
Multi-objective optimization [108]
Variable structure control [146]
Reinforcement learning [94]
Bayesian methods [147]
in automotive engineering applications. The following positions can be especially mentioned in this context:
For the selected topics eachDriver, Vehicle Dynamics

Control, Electric Vehicles, Driving Environmentthe
analysis has discovered a series of fuzzy application cases
of different nature that points to good versatile and flexible feasibility of fuzzy methods for automotive engineering tasks;
Relevant publication statistics indicates continuous
growth of fuzzy-related research activities in the selected
topics each;
Despite the currently available gap between theoretical
research and industrial implementations, a strong demand
in automotive systems with intelligent functions and involving more intuitive interaction with the driver and environment will stimulate a more wide adoption of fuzzy
tools and systems on serial vehicles.
The most feasible breakthrough in further development of

fuzzy systems and tools for automotive engineering applications can be expected from two kinds of Bfusion^. From theoretical side, the fusion of fuzzy methods with other methods
of soft computing or nonlinear control opens very promising
prospects. From technological side, the fusion of many vehicle controllers and estimators on common fuzzy basis can get
over numerical and linguistic uncertainties accompanying
complex processes of vehicle dynamics and vehicle-driverenvironment interaction. These trends would encourage the
researchers and engineers in seeking for novel applicative
fuzzy solutions in automotive engineering.
Page 7 of 10 29
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
Open Access This article is distributed under the terms of the
Creative Commons Attribution 4.0 International License (http://
creativecommons.org/licenses/by/4.0/), which permits unrestricted
use, distribution, and reproduction in any medium, provided you give
appropriate credit to the original author(s) and the source, provide a link
to the Creative Commons license, and indicate if changes were made.
21.
22.
23.
References
Pappis CP, Mamdani EH (1977) A fuzzy logic controller for a
traffic junction. IEEE Trans Syst Man Cybernet SMC-7(10):
707717
2. Sugeno M, Murakami K (1984) Fuzzy parking control of model
car, Proc. IEEE Conf Decis Control, pp. 902903
3. Sugeno M, Nishida M (1985) Fuzzy control of model car. Fuzzy
Sets Syst 16:103113
4. Ivanov V (2001) Fuzzy methods in ground vehicle engineering :
state-of-the-art and advanced applications, in Proc. 8th
International Conference on Structural Dynamics, EURODYN
2011, Leuven, Belgium, paper MS19-270
5. Kramer U, Rohr G (1982) A model of driver behaviour.
Ergonomics 25(10):891907
24.
1.
25.
26.
27.
28.
Kramer U (1985) On the application of fuzzy sets to the analysis of

the system driver-vehicle-environment. Automatica 21(1):101107
Nagai M, Kojima Y, Sato H (1987) Drivers decision and control
process during right-turning at intersections (application of fuzzy
linguistic model for accident analysis), SAE Technical Papers
Series, paper no. 871250
Kageyama I, Pacejka HB (1991) On a new driver model with
fuzzy control. Veh Syst Dyn 20(suppl 1):314324
Ehara N, Suzuki K (1991) Estimation of human reasoning state in
machine operation using fuzzy linear regression. Trans Jpn Soc
Mech Eng Part C 57:39073912
Lin C-T, Tsai S-F, Ko L-W (2013) EEG-based learning system for
online motion sickness level estimation in a dynamic vehicle environment. IEEE Trans Neural Netw Learn Syst 24(10):1689
1700, art. no. 6587600
Kuhn K-P, Heidinger A (1997) On-line driver type classification.
Int J Veh Des 18(6):616625
Lee D, Donnell ET (2007) Analysis of nighttime driver behavior
and pavement marking effects using fuzzy inference system. J
Comput Civ Eng 21(3):200210
Wu J-D, Chen T-R (2008) Development of a drowsiness warning
system based on the fuzzy logic images analysis. Expert Syst Appl
34(2):15561561
Boyraz P, Acar M, Kerr D (2008) Multi-sensor driver drowsiness
monitoring. Proc IMechE Part D J Automob Eng 222(11):18571878
Wahab A, Quek C, Tan CK, Takeda K (2009) Driving profile
modeling and recognition based on soft computing approach.
IEEE Trans Neural Netw 20(4):563582
Maeda M (1995) Fuzzy drive expert system for an automobile. Inf
Sci Appl 4(1):2948
Willumeit H-P (1997) Driver models: how to avoid critical yaw
movements of vehicles. Int J Veh Des 18(6):652667
Nigro J-M, Loriette-Rougegrez S, Rombaut M (2002) Driving
situation recognition with uncertainty management and rulebased systems. Eng Appl Artif Intell 15(34):217228
Hong C-W, Shio T-W (1996) Fuzzy control strategy design for an
autopilot on automobile chassis dynamometer test stands.
Mechatronics 6(5):537555
Hong C-W (1997) Tuning the fuzzy control autopilot strategy for
driving pattern simulation of road vehicles. Int J Veh Des 18(1):3552
Lee MC, Park MK, Yoo WS, Son K, Han MC (2003) Sliding
mode control of real-time PNU vehicle driving simulator and its
performance eveluation. JSME Int J Ser C Mech Syst Mach Elem
Manuf 46(2):557564
Takahashi H, Ukishima D, Kawamoto K, Hirota K (2007) A study
on predicting hazard factors for safe driving. IEEE0 Trans Ind
Electron 54(2):781789
Matthews ND, An PE, Roberts JM, Harris CJ (1998) A
neurofuzzy approach to future intelligent driver support systems.
Proc IMechE Part D J Automob Eng 212(1):4358
Takahashi H, Kuroda K (1999) Intelligent vehicle control considering drivers visual perception, Proc. IEEE Conference on
Intelligent Transportation Systems, pp. 252257
Chen G, Zhang W-G, Zhang X-N (2013) Fuzzy neural control for
unmanned robot applied to automotive test. Ind Robot 40(5):450
461, art. no. 17094477
Laurence P, Basset M, Coutant P, Gissinger G (2000) Lateral vehicle behaviour: comparison of subjective/objective assessment
using the Choquet integral. Veh Syst Dyn 34(5):357379
Ma X, Andrasson I (2007) Statistical analysis of driver behavior
data in different regimes of the car-following stage. Transportation
Research Record, no. 2018, pp. 8796
Moridpour S, Rose G, Sarvi M (2009) Modelling the heavy vehicle drivers lane changing decision under heavy traffic conditions.
Road Transp Res 18(4):4957
29 Page 8 of 10
29.
Hurwitz DS, Wang H, Knodler MA Jr, Ni D, Moore D (2012)

Fuzzy sets to describe driver behavior in the dilemma zone of
high-speed signalized intersections. Transport Res F: Traffic
Psychol Behav 15(2):132143
30. Vangi D, Virga A (2003) Evaluation of energy-saving driving
styles for bus drivers. Proc IMechE Part D J Automob Eng
217(4):299305
31. Syed FU, Filev D, Ying H (2007) Fuzzy rule-based driver advisory system for fuel economy improvement in a hybrid electric vehicle, Proc. of Annual Conference of the NAFIPS, no. 4271056,
pp. 178183
32. Luu HT, Nouvelire L, Mammar S (2010) Ecological and safe
driving assistance system: Design and strategy, Proc. of IEEE
Intelligent Vehicles Symposium, no. 5548012, pp. 129134
33. Stumpf H, Arendt G, Lux F (1988) Linking evaluation of subjective handling tests on the road with objectively measured data by
using fuzzy-set-theory, SAE Technical Papers Series, paper no.
885006
34. Madau DP, Yuan F, Davis Jr LI, Feldkamp LA (1993) Fuzzy logic
anti-lock brake system for a limited range coefficient of friction
surface, 1993 I.E. International Conference on Fuzzy Systems, pp.
883888
35. Gullett C, Elting D, Kowalczyk R, Fennich M, Hellenthal B
(1995) Intel fuzzy logic tool simplifies ABS design, Intel
Corporation Application Note, No. 272595001, pp. 18
36. Eichfeld H, Mertens A, Brandmeier T, Waidelich F, Graumann R,
Schwab M (1998) Applications of SAE 81C99x fuzzy coprocessors. Proc IEEE Int Conf Fuzzy Syst 1:1924
37. Mauer GF (1995) Fuzzy logic controller for an ABS braking system. IEEE Trans Fuzzy Syst 3(4):381388
38. Chen F-W, Liao T-L (2000) Nonlinear linearization controller and
genetic algorithm-based fuzzy logic controller for ABS systems
and their comparison. Int J Veh Des 24(4):334349
39. Kim S, Kim H, Sung Y-G (2001) Electronic control of braking
force distribution for vehicles using a direct adaptive fuzzy controller. KSME Int J 15(1):6680
40. Kuti I (2004) An adaptive fuzzy control to antilock non-straight
line braking using three dimensional (finite element) vehicle model. Int J Veh Des 34(4):327339
41. Klaus K, Hasemann M (1995) An embedded fuzzy anti-slippage
system for heavy duty off road vehicles. Inf Sci Appl 4(1):127
42. Hiller M, Schmitz T, Schuster C (1996) Design of a fuzzy traction
control system using spatial vehicle dynamics simulation. Heavy
Veh Syst 3(12):240248
43. Trebi-Ollennu A, Dolan JM, Khosla PK (2001) Adaptive fuzzy
throttle control for an all-terrain vehicle. Proc IMechE Part I J Syst
Control Eng 215(3):189198
44. Hessburg T, Tomizuka M (1995) An adaptation method for fuzzy
logic controllers in lateral vehicle guidance. Mechatronics 5(8):
873898
45. Boada BK, Boada MJL, Daz V (2005) Fuzzy-logic applied to
yaw moment control for vehicle stability. Veh Syst Dyn 43(10):
753770
46. Chadli M, El Hajjaji A, Oudghiri M (2008) Robust output fuzzy
control for vehicle lateral dynamic stability improvement. Int J
Model Identif Control 3(3):247257
47. Du H, Zhang N, Smith W (2009) Robust yaw moment control for
vehicle handling and stability. SAE Int J Passenger Cars Mech
Syst 2(1):772779
48. Ivanov V, Shyrokau B, Augsburg K, Gramstat S (2010)
Advancement of vehicle dynamics control with monitoring the
tire rolling environment. SAE Int J Passenger Cars Mech Syst
3(1):199216
49. Szosland A (2000) Fuzzy logic approach to four-wheel steering of
motor vehicle. Int J Veh Des 24(4):350359

50.
Huh K, Kim J (2001) Active steering control based on the estimated tire forces. J Dyn Syst Meas Control Trans ASME 123(3):
505511
51. Jalali K, Uchida T, McPhee J, Lambert S (2013) Development of
an advanced fuzzy active steering controller and a novel method to
tune the fuzzy controller. SAE Int J Passenger Cars Electron Electr
Syst 6(1):241254
52. Basset M, Zimmer C, Gissinger GL (1997) Fuzzy approach to the
real time longitudinal velocity estimation of a FWD car in critical
situations. Veh Syst Dyn 27(56):477489
53. Hwang JK, Song CK (2005) Fuzzy estimation of vehicle speed
using an accelerometer and wheel sensors. Int J Automot Technol
6(4):359365
54. Cheli F, Sabbioni E, Pesce M, Melzi S (2007) A methodology for
vehicle sideslip angle identification: comparison with experimental data. Veh Syst Dyn 45(6):549563
55. Zeyada YF, Elbeheiry EM, Elarabi ME (2000) Fault diagnosis and
-estimation for integrated chassis control of vehicles in emergency maneuvers. J Eng Appl Sci 47(4):665681
56. Yoshimura T, Emoto Y (2003) Steering and suspension system of
a full car model using fuzzy reasoning and disturbance observers.
Int J Veh Auton Syst 1(34):363386
57. March C, Shim T (2007) Integrated control of suspension and
front steering to enhance vehicle handling. Proc IMechE Part D
J Automob Eng 221(4):377391
58. Song J (2012) Integrated control of brake pressure and rear-wheel
steering to improve lateral stability with fuzzy logic. Int J Automot
Technol 13(4):563570
59. Ergun M (2007) Estimation of friction coefficient of asphalt concrete road surfaces using the fuzzy logic approach. Can J Civ Eng
34(9):11101118
60. Shyrokau BN, Ivanov VG (2008) Alterable fuzzy sets in automotive control applications. Int J Model Identif Control 3(3):305317
61. Li L, Song J, Li H-Z, Shan D-S, Kong L, Yang CC (2009)
Comprehensive prediction method of road friction for vehicle dynamics control. Proc IMechE Part D J Automob Eng 223(8):9871002
62. Ivanov V, Shyrokau B, Augsburg K, Algin V (2010) Fuzzy evaluation of tyre-surface interaction parameters. J Terrramech 47(2):
113130
63. Ivanov V (2011) Fuzzy architecture of systems with alterable information: case study for tyreground friction estimators. Int J
Reliab Saf 5(3/4):398419
64. Yester JL, McFall RH (1992) Fuzzy logic controller for active
suspension. Proc Soc Automot Eng P-260:259270
65. Lin Y-J, Lu Y-Q (1993) Toward better ride performance of vehicle
suspension systems via intelligent control. Am Soc Mech Eng
Dyn Syste Control Div (Publ) DSC 48:105112
66. Jianghong Z, Long T (1994) An evaluation of comfort of a bus
seat. Appl Ergon 25(6):386392
67. Eslaminasab N, Biglarbegian M, Melek WW, Golnaraghi MF
(2007) A neural network based fuzzy control approach to improve
ride comfort and road handling of heavy vehicles using semiactive dampers. Int J Heavy Veh Syst 14(2):135157
68. Suzuki T, Takahashi M (2010) Robust active suspension control
for vibration reduction of passengers body. Trans JSME Part C
76(762):380387
69. Cherry AS, Jones RP (1995) Fuzzy logic control of an automotive
suspension system. IEE Proc Control Theory Appl 142(2):149160
70. Yoshimura T, Isari Y, Li Q, Hino J (1997) Active suspension of
motor coaches using skyhook damper and fuzzy logic control.
Control Eng Pract 5(2):175184
71. Caponetto R, Diamante O, Fargione G, Risitano A, Tringali D
(2003) A soft computing approach to fuzzy sky-hook control of
semiactive suspension. IEEE Trans Control Syst Technol 11(6):
786798

72.
Sie W-T, Lian R-J, Lin B-F (2006) Enhancing grey prediction
fuzzy controller for active suspension systems. Veh Syst Dyn
44(5):407430
73. Kaldas MMS, alikan K, Henze R, Kkay F (2013) Rule
optimized fuzzy logic controller for full vehicle semi-active suspension. SAE Int J Passenger Cars Mech Syst 6(1):332344
74. Choi S-B, Park D-W, Suh M-S (2002) Fuzzy sky-ground hook
control of a tracked vehicle featuring semi-active
electrorheological suspension units. J Dyn Syst Meas Control
Trans ASME 124(1):150157
75. Giuclea M, Sireteanu T, Stancioiu D, Stammers CW (2004) Model
parameter identification for vehicle vibration control with
magnetorheological dampers using computational intelligence
methods. Proc IMechE Part I J Syst Control Eng 218(7):569581
76. Zheng L, Li YN, Baz A (2009) Fuzzy-sliding mode control of a
full car semi-active suspension systems with MR dampers. Smart
Struct Syst 5(3):261277
77. Lee S, Kim W-J (2010) Active suspension control with directdrive tubular linear brushless permanent-magnet motor. IEEE
Trans Control Syst Technol 18(4):859870, art. no. 5286231
78. Hogle RA, Bonissone PP (1994) Fuzzy algorithm for path selection in autonomous vehicle navigation. Proc IEEE Conf Decis
Control:898900
79. Freisleben B, Kunkelmann T (1993) Combining fuzzy logic and
neural networks to control an autonomous vehicle, 1993 I.E.
International Conference on Fuzzy Systems, pp. 321326
80. Blchl B, Tsinas L (1994) Automatic road following using fuzzy
control. Control Eng Pract 2(2):305311
81. Chakroborty P, Kikuchi S (1999) Evaluation of the general motors
based car-following models and a proposed fuzzy inference model. Transp Res Part C Emerg Technol 7(4):209235
82. Naranjo JE, Gonzlez C, Reviejo J, Garca R, De Pedro T (2003)
Adaptive fuzzy control for inter-vehicle gap keeping. IEEE Trans
Intell Transp Syst 4(3):132142
83. Qiao MS, Takeda H (1995) Learning algorithm of environmental
recognition in driving vehicle. IEEE Trans Syst Man Cybernet
25(6):917925
84. Stover JA, Hall DL, Gibson RE (1996) A fuzzy-logic architecture
for autonomous multisensor data fusion. IEEE Trans Ind Electron
43(3):403410
85. Goebel KF, Agogino AM (2001) Sensor validation and fusion for
automated vehicle control using fuzzy techniques. J Dyn Syst
Meas Control Trans ASME 123(1):145146
86. Rigatos G, Tzafestas S (2007) Extended Kalman filtering for
fuzzy modelling and multi-sensor fusion. Math Comput Model
Dyn Syst 13(3):251266
87. Kim SY, Choi HC, Won WJ, Oh SY (2009) Driving environment
assessment using fusion of in- and out-of-vehicle vision systems.
Int J Automot Technol 10(1):103113
88. Broggi A, Cerri P, Ghidoni S, Grisleri P, Jung HG (2009) A new
approach to urban pedestrian detection for automatic braking.
IEEE Trans Intell Transp Syst 10(4):594605, art. no. 5290131
89. Filev DP, Kolmanovsky I (2012) Markov chain modeling and onboard identification for automotive vehicles. Lect Notes Control
Inf Sci 418:111128
90. Bauer M, Tomizuka M (1996) Fuzzy logic traction controllers and
their effect on longitudinal vehicle platoon systems. Veh Syst Dyn
25(4):277303
91. Lee GD, Kim SW, Yim YU, Jung JH, Oh SY, Kim BS (1999)
Longitudinal and lateral control system development for a platoon
of vehicles, IEEE Conference on Intelligent Transportation
Systems, Proceedings, ITSC, pp. 605610
92. Kim K, Kim J, Huh KS, Yi K, Cho D (2006) A real-time multivehicle simulator for longitudinal controller design. Veh Syst Dyn
44(5):369386
Page 9 of 10 29
93.
94.
95.
96.
97.
98.
99.
100.
101.
102.
103.
104.
105.
106.
107.
108.
109.
110.
111.
112.
113.
Pasquier M, Quek C, Toh M (2001) Fuzzylot: a novel selforganising fuzzy-neural rule-based pilot system for automated vehicles. Neural Netw 14(8):10991112
Dai X, Li CK, Rad AB (2005) An approach to tune fuzzy controllers based on reinforcement learning for autonomous vehicle control. IEEE Trans Intell Transp Syst 6(3):285293
Milans V, Villagr J, Prez J, Gonzlez C (2012) Low-speed
longitudinal controllers for mass-produced cars: a comparative
study. IEEE Trans Ind Electron 59(1):620628, art. no. 5759771
Prez J, Milans V, Godoy J, Villagr J, Onieva E (2013)
Cooperative controllers for highways based on human experience.
Expert Syst Appl 40(4):10241033
Kim HM, Dickerson J, Kosko B (1996) Fuzzy throttle and brake
control for platoons of smart cars. Fuzzy Sets Syst 84(3):209234
Naranjo JE, Reviejo J, Gonzlez C, Garca R, De Pedro T (2004)
A throttle and brake fuzzy controller: towards the automatic car.
Lect Notes Comput Sci (Incl Subseries Lec Notes Artif Intell Lect
Notes Bioinforma 2809:291301
Naranjo JE, Gonzlez C, Garca R, De Pedro T, Haber RE (2005)
Power-steering control architecture for automatic driving. IEEE
Trans Intell Transp Syst 6(4):406415
Ryu Y-W, Oh S-Y, Kim S-Y (2008) Robust automatic parking
without odometry using an evolutionary fuzzy logic controller.
Int J Control Autom Syst 6(3):434443
Chen C-Y, Feng H-M (2009) Hybrid intelligent vision-based carlike vehicle backing systems design. Expert Syst Appl 36(4):
75007509
Shiao Y, Yu C-J (2012) Automatic parking control for vehicles
with stereo vision. Adv Sci Lett 8:4954
Farrall SD, Jones RP (1993) Energy management in an automotive electric/heat engine hybrid powertrain using fuzzy decision
making, Proc. 1993 International Symposium on Intelligent
Control, pp. 463468
Cerruto E, Consoli A, Raciti A, Testa A (1994) Fuzzy logic based
efficiency improvement of an urban electric vehicle. IECON Proc
(Ind Electron Conf 2:13041309
Vermeiren L, Guerra TM, Paganelli G (1997) Application of fuzzy
set theory to electric car control. Cybern Syst 28(8):675693
Baumann BM, Washington G, Glenn BC, Rizzoni G (2000)
Mechatronic design and control of hybrid electric vehicles.
IEEE/ASME Trans Mechatron 5(1):5872
Schouten NJ, Salman MA, Kheir NA (2002) Fuzzy logic control
for parallel hybrid vehicles. IEEE Trans Control Syst Technol
10(3):460468
Ippolito L, Loia V, Siano P (2003) Extended fuzzy C-means and
genetic algorithms to optimize power flow management in hybrid
electric vehicles. Fuzzy Optim Decis Making 2(4):359374
Ahn H-S, Lee NS, Moon CW, Jeong G-M (2007) Fuel economy
improvement for fuel cell hybrid electric vehicles using fuzzy
logic-based power distribution control. Int J Automot Technol
8(5):651658
Solano Martnez J, Mulot J, Harel F, Hissel D, Pra M-C, John RI,
Amiet M (2013) Experimental validation of a type-2 fuzzy logic
controller for energy management in hybrid electrical vehicles.
Eng Appl Artif Intell 26(7):17721779
Yoshimura T, Uchida H, Nasu H, Hino J, Ueno R (1997) Traction
force control of an electric vehicle in 2WS-4WD mode using fuzzy
reasoning. Int J Veh Des 18(5):442454
Pusca R, Ait-Amirat Y, Berthon A, Kauffmann JM (2002) Fuzzy
logic based control for electric vehicle with four separate traction
drives. Proc IEEE Veh Technol Conf 4:20892097
Kim D-H, Kim J-M, Hwang S-H, Kim H-S (2007) Optimal brake
torque distribution for a four-wheel-drive hybrid electric vehicle
stability enhancement. Proc IMechE Part D J Automob Eng
221(11):13571366
29 Page 10 of 10
114.
115.
116.
117.
118.
119.
120.
121.
122.
123.
124.
125.
126.
127.
128.
129.
130.
Huang H (1999) Electrical two-speed propulsion by motor winding switching and its control strategies for electric vehicles. IEEE
Trans Veh Technol 48(2):607618
Sepe RB Jr, Miller JM (2002) Real-time collaborative experimentation via the internet: fuzzy efficiency optimization of a hybrid
electric vehicle starter/alternator. Int J Veh Auton Szstems 1(1):
6382
Ferreira AA, Pomilio JA, Spiazzi G, de Araujo Silva L (2008)
Energy management fuzzy logic supervisory for electric vehicle
power supplies system. IEEE Trans Power Electron 23(1):107
115
Quigley CP, Ball RJ, Jones RP (2000) Fuzzy modelling approach
to the prediction of journey parameters for hybrid electric vehicle
control. Proc IMechE Part D J Automob Eng 214(8):875885
Cunningham I, Burnham K (2013) Online use of the fuzzy transform in the estimation of electric vehicle range. Meas Control
(UK) 46(9):277282
Chau KT, Wu KC, Chan CC, Shen WX (2003) A new battery
capacity indicator for nickel-metal hydride battery powered electric vehicles using adaptive neuro-fuzzy inference system. Energy
Convers Manag 44(13):20592071
Albertus P, Couts J, Srinivasan V, Newman J (2008) A combined
model for determining capacity usage and battery size for hybrid
and plug-in hybrid electric vehicles. J Power Sources 183(2):771
782
Fleischer C, Waag W, Bai Z, Sauer DU (2013) On-line self-learning time forward voltage prognosis for lithium-ion batteries using
adaptive neuro-fuzzy inference system. J Power Sources 243:728
749
Hayashi K, Shimizu Y, Dote Y, Takayama A, Hirako A (1995)
Neuro fuzzy transmission control for automobile with variable
loads. IEEE Trans Control Syst Technol 3(1):4952
Halfmann C, Ayoubi M, Holzmann H (1997) Supervision of vehicles tyre pressures by measurement of body accelerations.
Control Eng Pract 5(8):11511159
Hissel D, Candusso D, Harel F (2007) Fuzzy-clustering durability
diagnosis of polymer electrolyte fuel cells dedicated to transportation applications. IEEE Trans Veh Technol 56(5 I):24142420,
art. no. 4305501
Kim MH, Lee S, Lee KC (2011) A fuzzy predictive redundancy
system for fault-tolerance of x-by-wire systems. Microprocess
Microsyst 35(5):453461
Mur S, Demichela M, Piccinini N (2006) Assessment of the risk
of occupational accidents using a Bfuzzy^ approach. Cognit
Technol Work 8(2):103112
Mon Y-J (2007) Airbag controller designed by adaptive-networkbased fuzzy inference system (ANFIS). Fuzzy Sets Syst 158(24):
27062714
Khayyam H, Nahavandi S, Hu E, Kouzani A, Chonka A, Abawajy
J, Marano V, Davis S (2011) Intelligent energy management control of vehicle air conditioning via look-ahead system. Appl
Therm Eng 31(16):31473160
Nopiah ZM, Junoh AK, Ariffin AK (2013) Noise annoyance
fuzzy index in passenger car cabin. Int J Veh Noise Vib 9(34):
216233
Farkas L, Moens D, Donders S, Vandepitte D (2012) Optimisation
study of a vehicle bumper subsystem with fuzzy parameters. Mech
Syst Signal Process 32:5960

131.
132.
133.
134.
135.
136.
137.
138.
139.
140.
141.
142.
143.
144.
145.
146.
147.
Filev D, Lu J, Tseng F, Prakah-Asante K (2011) Real-time driver

characterization during car following using stochastic evolving
models, IEEE International Conference on Systems, Man and
Cybernetics, pp. 10311036
Lu J, Filev D, Tseng F (2015) Real-time determination of drivers
driving behavior during car following. SAE Int J Passenger Cars
Electron Electr Syst, vol. 8, no. 2
D. Filev, F. Tseng, J. Kristinsson, and R. McGee, Real-time determination of drivers driving behavior during car following, IEEE
Symposium on Computational Intelligence in Vehicles and
Transportaion Systems, pp. 8791, 2011
Tango F, Minin L, Tesauri F, Montanari R (2010) Field tests and
machine learning approaches for refining algorithms and correlations of drivers model parameters. Appl Ergon 41(2):211224
Dubart D, Loslever P, Popieul JC, Moessinger M (2011) Multiple
correspondence analysis of fuzzyfied task performance and
psycho-physiological test data: Use in real car following situations, 12th IFAC, IFIP, IFORS, IEA Symposium on Analysis,
Design, and Evaluation of Human-Machine Systems
Gusikhin O, Filev D, Rychtyckyi N (2008) Intelligent vehicle
system: Applications and new trends, Lecture Notes in Electrical
Engineering, col. 15, pp. 314
Dahmani H, Chadli M, Rabhi A, El Hajjaji A (2013) Vehicle
dynamic estimation with road bank angle consideration for rollover detection: Theoretical and experimental studies. Veh Syst
Dyn 51(12):18531871
Yeh EC, Tsao YJ (1994) Fuzzy preview control scheme of active
suspension for rough road. Int J Veh Des 15(12):166180
Cheok KC, Smid GE, McCune DJ (2000) Multisensor-based collision avoidance system with application to a military HMMWV,
IEEE Conference on Intelligent Transportation Systems,
Proceedings, ITSC, pp. 288292
Economou JT, Colyer RE (2006) Fuzzy-hybrid modelling of an
Ackerman steered electric vehicle. Int J Approx Reason 41(3):
343368
Yang JH, Yang CC (2008) Adaptive sliding mode control for
vehicle brake systems with fuzzy composition. Int J Model
Identif Control 4(2):112126
Chou J-H, Chen S-H, Lee F-Z (1998) Grey-fuzzy control for active suspension design. Int J Veh Des 19(1):6577
Li Y, Ling Z, Liu Y, Qiao Y (2005) Method of fuzzy-PID control
on vehicle longitudinal dynamics system, Lecture Notes in
Artificial Intelligence, no. 3613 (PART I), pp. 822832
Huang S, Ren W (1999) Use of neural fuzzy networks with mixed
genetic/gradient algorithm in automated vehicle control. IEEE
Trans Ind Electron 46(6):10901102
Kataoka H, Sado H, Sakai I, Hori Y (2001) Optimal slip ratio
estimator for traction control system of electric vehicle based on
fuzzy inference. Electr Eng Jpn (Engl Transl Denki Gakkai
Ronbunshi) 135(3):5663
Guglielmino E, Stammers CW, Stncioiu D, Sireteanu T,
Ghigliazza R (2005) Hybrid variable structure-fuzzy control of a
magnetorheological damper for a seat suspension. Int J Veh Auton
Syst 3(1):3446
Senaratne R, Hardy D, Vanderaa B, Halgamuge S (2007) Driver
fatigue detection by fusing multiple cues, Lecture Notes in
Computer Science, 4492 LNCS (PART 2), pp. 801809
Biol Cybern (2016) 110:91116

DOI 10.1007/s00422-016-0682-x
REVIEW
A review of human sensory dynamics for application to models

of driver steering and speed control
Christopher J. Nash1 David J. Cole1
Robert S. Bigler1
Received: 13 November 2015 / Accepted: 22 February 2016 / Published online: 16 April 2016
The Author(s) 2016. This article is published with open access at Springerlink.com
Abstract In comparison with the high level of knowledge

about vehicle dynamics which exists nowadays, the role of
the driver in the drivervehicle system is still relatively poorly
understood. A large variety of driver models exist for various applications; however, few of them take account of the
drivers sensory dynamics, and those that do are limited in
their scope and accuracy. A review of the literature has been
carried out to consolidate information from previous studies which may be useful when incorporating human sensory
systems into the design of a driver model. This includes
information on sensory dynamics, delays, thresholds and
integration of multiple sensory stimuli. This review should
provide a basis for further study into sensory perception during driving.
Keywords Sensory dynamics Driver modelling
Perception thresholds Sensory integration Drivervehicle
dynamics
1 Introduction
The continued development of advanced driver assistance
systems (ADAS) in road vehicles is resulting in increasingly
complex interactions between driver and vehicle (Gordon
This work was supported by the UK Engineering and Physical
Sciences Research Council (EP/P505445/1) (studentship for Nash).
Christopher J. Nash
cn320@cam.ac.uk
David J. Cole
djc13@cam.ac.uk
Cambridge University Engineering Department, Trumpington

Street, Cambridge CB2 1PZ, UK
and Lidberg 2015). However, the role of the human driver

in controlling the vehicle is still poorly understood. Consequently the vehicle development process still relies heavily
on subjective evaluation of prototype vehicles by test drivers,
which is expensive and time consuming. By building a deeper
understanding of the interactions between driver and vehicle,
models can be developed to assist with the design and evaluation of vehicle components and systems. One feature of
drivervehicle control that has been neglected to date is the
sensory perception of the driver. The aim of this paper is
to review the role of human sensory systems in the driving
task, with a view to improving the capability of mathematical
models of the driver.
Driving a vehicle involves a wide range of information
processing levels, from the high-level navigation task to the
low-level control of vehicle speed and direction. The focus of
this review is on the role of human sensory dynamics in the
low-level control task. Donges (1978) considered the steering control task as the superposition of a target following
task (feedforward control) and a disturbance rejection task
(feedback control). Disturbances may act on the vehicle from
sources such as wind gusts, uneven road surfaces and nonlinearities in the vehicle dynamics, or they may originate from
the driver due to physiological noise sources, constraints and
nonlinearities.
A simplified block diagram of the feedforward and feedback control of vehicle direction and speed is shown in Fig. 1.
The driver previews the future road path using their visual
system and then, using an internal model of the vehicle
dynamics, determines target path and speed profiles and corresponding feedforward control actions (Timings and Cole
2013, 2014). Simultaneously, the driver senses the motion
of the vehicle in relation to the target profiles and generates
feedback control actions to reduce the effect of disturbances.
The hypothesis presented in Fig. 1 assumes that feedback of
123
92
Biol Cybern (2016) 110:91116

Driver
noise
Road
geometry
Driver
Road
preview
Preview
time step
Trajectory
planning
Steering
control
Motion
perception
Speed
control
Vehicle
motion
Steering
wheel angle
Pedal
forces
Preview horizon
Fig. 2 Preview model of drivers perception of the upcoming road

path, used for feedforward steering and speed control (Sharp and Valtetsiotis 2001). The driver looks straight ahead and takes a series of
measurements of the lateral displacement of the target path up to the
preview horizon
Vehicle
Vehicle
disturbances
Fig. 1 Block diagram of tasks carried out while driving. The driver
must use their perceived information about the motion of the vehicle
and the upcoming road geometry to plan a desired trajectory and then
calculate the required steering wheel angle and pedal forces to achieve
this trajectory as closely as possible
vehicle motion is not used directly for generating the feedforward control action; however, the feedback loop is able to
correct for any discrepancies introduced by imperfections in
the drivers feedforward control. It has been found that without visual feedback during lane change or obstacle avoidance
manoeuvres drivers do not always initiate the return phase of
the manoeuvre, failing to steer back towards the target path
(Wallis et al. 2002; Cloete and Wallis 2009).
Modelling the driver mathematically has been the subject of research for many decades. Comprehensive reviews
are provided by Macadam (2003) and Plchl and Edelmann
(2007). Recent research has focussed on the application
of optimal control theory, using model predictive or linear
quadratic controllers that are able to preview the future road
path, as shown in Fig. 2, and calculate an optimal sequence
of control actions (Macadam 1981; Sharp and Valtetsiotis
2001; Peng 2002; Cole et al. 2006). This approach has been
extended to include neuromuscular dynamics (Pick and Cole
2007, 2008; Odhams and Cole 2009; Abbink et al. 2011;
Cole 2012) and to the control of nonlinear vehicle dynamics
(Ungoren and Peng 2005; Thommyppillai et al. 2009; Keen
and Cole 2011). Feedforward and feedback control are usually assumed to share a common objective function. Timings
and Cole (2014) synthesised independent feedforward and
feedback controllers to examine in more detail the robustness of the drivers control strategy to disturbances.
123
Previewed
displacements
While driver steering control has a fairly well-defined

objective, to follow a target line and stay within road boundaries, the motivation for drivers speed choice depends on
the situation. In a normal driving situation drivers will balance factors such as safety, comfort, journey time and control
effort (Prokop 2001; Odhams and Cole 2004). Drivers have
been found to decrease their speed to minimise their lateral acceleration in corners (Ritchie et al. 1968; Herrin and
Neuhardt 1974; Reymond et al. 2001). Road width has also
been found to affect speed choice, with drivers adjusting
their speed to remain within lane boundaries (Bottoms 1983;
Defazio et al. 1992). In contrast, racing drivers aim to maximise their lateral acceleration within the limits of the tyres
in order to minimise lap time (Timings and Cole 2014; Lot
and Dal Bianco 2015). In situations with heavy traffic, driver
speed choice may also be dictated by the speed of other vehicles, with the driver aiming to maintain a safe distance behind
the car in front (Boer 1999; Kondoh et al. 2008).
Despite these developments, most models assume the
driver has full knowledge of the vehicle states, and no existing
driver models appear to take full advantage of current understanding of human sensory dynamics. While this review is
primarily focussed on driving of road vehicles, clear parallels can be drawn with research into pilots in the aerospace
industry. Indeed, sensory dynamics have been considered in
greater detail in this area, and many of the studies cited in this
review have come from work carried out by aerospace engineers to investigate human perception during control tasks.
In particular, models of sensory dynamics have been used
in studies carried out in flight simulators to understand how
sensory information is used during real and simulated flight
(Pool et al. 2008; Ellerbroek et al. 2008; Nieuwenhuizen et al.
2013; Drop et al. 2013; Zaal et al. 2009a,c, 2010, 2012, 2013).
Driving is just one of many human sensorimotor tasks that
involve perceiving stimuli in the surrounding environment
and responding with a physical action. The neurophysiological processes involved in such tasks are shown in Fig. 3. A
stimulus may excite various senses, which produce chemical
Biol Cybern (2016) 110:91116
93
Receptor
nonlinearities
Receptor
dynamics
Stimulus
= Delay sources
+
+
Sensors
Nerve
conduction
= Noise sources
Brain
Signal
processing
Spontaneous
neuron firing
Reflex
response
Sensory
integration
Muscles
+
Response
Neuromuscular
dynamics
Muscle
activation
Response
planning
Internal
model
+
Nerves
Neuromuscular
nonlinearities
Internal
model
errors
+
+
Fig. 3 Diagram of the main processes carried out by the sensorimotor

system to generate a physical response to a sensory stimulus. Stimuli are
perceived by various sensors, which transmit electrical signals through
the nerves to the brain. The brain processes and integrates these sensory signals and then plans a response using an internal model of the
environment. The planned signals are sent to the muscles and shaped
by the neuromuscular dynamics to give a physical response. There are
various sources of time delays, shown by boxes, and noise, shown by
ovals
signals characterised by the dynamics of the sensory receptors (explored in Sect. 2). Sensory signals are then transmitted
through the nerves as electrical impulses caused by firing
neurons, with the firing rate encoding a frequency-modulated
signal (Carpenter 1984). Certain stimuli can elicit reflexive
responses which bypass the brain by activating motor neurons emerging from the spinal cord (Carpenter 1984).
There are physical and biochemical limitations to the
speed with which each of the processes shown in boxes in
Fig. 3 can be carried out; therefore, time delays are introduced
into the sensorimotor system. These delays are discussed
further in Sect. 3. In addition, noise is introduced due to
nonlinearities in the receptor and neuromuscular dynamics,
errors in the brains internal models and spontaneous firing
of neurons (Fernandez and Goldberg 1971). This means that
humans are unable to measure stimuli with perfect accuracy
or plan and execute an ideal response. It also results in thresholds below which stimuli cannot be perceived, as discussed
in Sect. 4.
Once the sensory signals are received in the brain, they
are processed in the sensory cortex in order to extract the
information from the encoded signals transmitted through
the nerves (Kandel et al. 2000). The information from the
different senses is then integrated to form a single representation of the surrounding environment, as explained further in
Sect. 5. Based on this, the physical response to the perceived
stimuli is planned using internal models of the human body
and the surrounding world (Wolpert and Ghahramani 2000).
The signals required to activate the muscles are generated in

the motor cortex and fine-tuned in the cerebellum using feedback from the sensory measurements (Kandel et al. 2000).
Signals are then transmitted along motor neurons which activate muscle fibres, causing them to contract. The physical
response is shaped by the dynamic properties of the activated
muscles. In the context of driving, earlier studies have measured and modelled the neuromuscular dynamics of drivers
arms holding a steering wheel (Pick and Cole 2007, 2008;
Odhams and Cole 2009; Cole 2012) and legs actuating a gas
pedal (Abbink et al. 2011).
An important feature of perception during driving tasks
is that the stimuli perceived by the drivers sensory systems
arise from the motion of the vehicle, which is controlled by
the driver. This means that the driver is involved in an active
closed-loop perception and control task, as opposed to a passenger who is a passive observer (Flach 1990). The driver is
able to anticipate future motion of the vehicle, allowing more
accurate sensory integration as discussed in Sect. 5. Driving
also involves many sensory stimuli being presented simultaneously in different axes and stimulating different sensors
(multimodal) compared with sensory measurements which
have been carried out in one axis to stimulate one sensor
(unimodal). Care must be taken when relating results from
investigations carried out in passive, unimodal conditions to
models of active, multimodal control and perception. This is
discussed in relation to time delays in Sect. 3 and sensory
thresholds in Sect. 4.
123
94
The scope of this review is broad, and thus it is not possible to review every topic in great detail; each section could be
extended significantly. However, the aim of the review is to
give an overview of the key results from the literature, with
particular focus on motivating and informing further development of driver models incorporating human sensory system
dynamics. Both steering and speed control are considered
concurrently, since in many cases the sensory mechanisms
discussed are relevant for both control tasks. The main findings of the review are summarised and discussed in Sect. 6.
The review extends considerably an earlier review by Bigler
and Cole (2011).
2 Sensory dynamics
Various sensory systems are used by the driver to infer the
state of the vehicle and its surroundings. The main sensory
systems used in the control of vehicle speed and direction
are:
Visual: The visual system is the only means the driver
has of detecting the upcoming road geometry. The visual
system can also sense the motion of the vehicle relative
to the surrounding environment.
Vestibular: The vestibular organs are located within the
inner ear, and they sense rotations and translations of the
drivers head.
Somatosensory: Somatosensors include a wide range of
sensory organs which detect various states of the body,
such as contact pressure, temperature, limb position and
pain. They include proprioceptors which detect joint
angles, muscle lengths and tensions and their derivatives.
The following subsections give an overview of the published
literature on these three sensory systems. Other senses such
as hearing may also play a role but will not be discussed in
detail.
2.1 Visual system
Visual perception is the subject of significant research activity in psychology, neuroscience and biology. There is still
much to understand about how a human interprets the neural
signals received by the retina from a potentially complex
three-dimensional visual scene containing objects that might
be familiar or unfamiliar, and moving or stationary, with
a moving or stationary observer. The various processes
involved in visual perception are discussed in detail by Gibson (1950), Johansson (1975), Ullman (1979), Nakayama
(1985), Lappe et al. (1999) and Raudies and Neumann
(2012). Human visual perception is a complex, multi-layered
process, and for the purpose of driver modelling it is not nec-
123
Biol Cybern (2016) 110:91116
essary or feasible to model all aspects. Therefore, the focus of

this review is on the most relevant results towards modelling
visual perception in a driving environment.
In the two-level model of vehicle control (Donges 1978),
the visual system is used in both the feedback task and the
feedforward task. The feedback task involves using the visual
system in combination with the vestibular and somatosensory systems to perceive the motion of the driver and thus
of the vehicle, which in turn is used to perform feedback
control of the vehicle. In the feedforward task, the visual
system views the geometry of the road ahead of the vehicle
so that feedforward control inputs to the vehicle can be generated. Higher levels of the driving task, not considered in
this review, involve the visual system in perceiving additional
information such as motion of other vehicles and pedestrians.
2.1.1 Perception of self-motion (feedback)
Visually induced motion perception is typically caused by
motion of the eyes relative to fixed surroundings, although
illusory self-motion perception known as vection can also
be induced by moving surroundings (Dichgans and Brandt
1978). Since vehicle motion is primarily planar, the role of
the drivers visual system in perceiving self-motion is mainly
concerned with three axes: longitudinal and lateral translations, and yaw (heading) rotations.
Various mechanisms have been suggested for visual
motion perception, such as optic flow (Gibson 1950; Koenderink 1986; Lappe et al. 1999). This is the velocity field
created as points in the visual scene flow over the retina,
along lines known as streamers. Optic flow patterns while
driving on straight and curved roads are shown by the dashed
lines in Fig. 4. For straight motion, the streamers all originate
from a point directly in front of the observer known as the
focus of radial outflow (FRO). This can be used as a visual
cue to control the vehicles heading direction (Gibson 1950),
for example by aligning with the vanishing point at the end
of a straight road. For rotational motion, the streamers are
curved and the FRO does not exist, although the point on the
horizon directly in front of the observer may still be used as a
visual cue to heading direction (Grunwald and Merhav 1976).
However, Riemersma (1981) suggested that the FRO and
heading direction are too crude to play a role in car driving.
Multi-level models of perception of motion from optic flow
have been developed (Grossberg et al. 2001; Mingolla 2003;
Browning et al. 2009); however, these descriptions do not
lead easily to a simple relationship between vehicle motion
and visually perceived motion, as they are dependent on the
characteristics of the surroundings.
Alternatively, it has been proposed that humans measure
the rates of change of vectors between themselves and specific objects in the visual field (Gordon 1965; Zacharias
et al. 1985). This allows drivers to calculate their time-to-
Biol Cybern (2016) 110:91116
Vanishing
point
95
FRO
Future
path point
Tangent
point
(a)
(b)
Fig. 4 Potential candidates for visual cues used while driving along a
a straight road and b a curved road. Optic flow patterns are shown by
dashed lines. Drivers have also been found to use other objects and the
road boundaries and centre line as visual references. Points such as the
tangent point and future path point have been identified as fixation
points for feedforward visual perception.
collision with objects, which can be particularly useful when

following a leading vehicle (Kondoh et al. 2008). The distance and relative velocity of the objects can only be inferred
with prior knowledge of the objects size or by comparison
between two visually similar environments (Gordon 1965;
Bremmer and Lappe 1999). Road edges and centre line have
also been identified as key visual features used by drivers
(Gordon 1966; Riemersma 1981).
Because of the variety of mechanisms involved in visual
perception, it is difficult to say what constitutes the input
to the visual system. Optic flow models would suggest that
velocities are measured, although the FRO can be used to
measure heading direction (yaw angle), and it is clearly possible to discriminate translational displacements with reference
to stationary features such as road markers. Gordon (1965)
used the unnatural appearance of the acceleration field to
argue that accelerations and higher derivatives are not directly
sensed by the visual system. The most appropriate inputs to
the feedback component of the drivers visual system therefore appear to be translational and angular velocities. Since
displacements and angles can only be measured with respect
to references such as road markers, they can be included
within models of drivers feedforward visual perception.
It is not clear from the mechanisms involved in visual perception whether the perceived rotational and translational
velocities depend on the frequency of the stimulus. One possible approach is simply to assume unity gains between the
actual and perceived velocities. An alternative estimate of
the frequency response of the visual system may be obtained
from sensory threshold measurements (Soyka et al. 2011,
2012, see Sect. 4 for more information). Riemersma (1981)
and Bigler (2013) both measured thresholds of visual perception of lateral and yaw velocities, superimposed on a
constant longitudinal velocity. Both studies presented subjects with a typical driving scene, with Riemersma (1981)
displaying edge lines for a straight road and Bigler (2013)
displaying a more realistic rendering of a straight road bordered by trees. Riemersma (1981) found that lateral and yaw
thresholds were independent of longitudinal speed. Bigler

(2013) found thresholds for stimuli of different frequencies,
and reanalysing the results using the model of (Soyka et al.
2011, 2012) gives visual dynamics that can be described by
a low-pass filter, given by:
Hvi (s) =
vi
s + vi
(1)
and taking lateral velocity and yaw angular velocity as inputs.

The same cutoff frequency vi = 0.810 rad/s was found to
fit the results for both sway and yaw motion. This low-pass
characteristic was also seen by Riemersma (1981). In the
absence of direct measurements of nervous responses to sensory stimulation, this model inferred from sensory threshold
data can be used to give some insight into the function of
the visual system. However, further research is needed to
validate this approach.
2.1.2 Perception of road path geometry (feedforward)
One of the key characteristics of driving tasks is the ability of the driver to use their visual system to preview
the road ahead in order to carry out feedforward control.
Studies have investigated the key features of road geometry
which are perceived while driving, often using eye tracking
instrumentation to investigate where the drivers look. Shinar et al. (1977) found a difference between straight roads,
where drivers tend to focus near the FRO, and curved roads,
where drivers scan the geometry of the curve. Many studies
have found that drivers focus on the tangent point on the
inside of a bend, as shown in Fig. 4 (Land and Lee 1994;
Boer 1996; Kandil et al. 2009, 2010). The angle between the
current vehicle heading vector and the tangent point can be
used to estimate the road curvature (Land and Lee 1994) and
required steering angle (Kandil et al. 2009). Other studies
have suggested drivers may look at a point on the predicted
vehicle path, the future path point (Land 1998) as shown in
Fig. 4. There is no overwhelming evidence in favour of the
123
96
tangent point over the future path point or other nearby points
as a fixation point during driving (Mars 2008; Robertshaw
and Wilkie 2008; Lappi et al. 2013).
Eye tracking studies have found that drivers tend to focus
on a point around 12 s ahead of the vehicle on straight roads
(Land and Lee 1994; Donges 1978), and that their gaze tends
to move to an upcoming curve around 1 s before they steer in
that direction (Chattington et al. 2007; Land and Tatler 2001).
Drivers have also been found to make short look-ahead fixations, looking further along the road for short periods of time
(Lehtonen et al. 2013). While eye tracking instrumentation is
useful for determining the gaze direction of a driver, Land and
Lee (1994) noted that it does not necessarily indicate where
the driver is directing their attention, because the driver may
be using their peripheral vision to gather information about
road geometry away from the gaze point. Grunwald and Merhav (1976) and Land and Horwood (1995) both measured
driver performance with only certain parts of the road visible
and found that the full visual control task can be represented
by two viewing points, one near to the driver and one further
down the road. Land and Horwood (1995) found that performance was not degraded from the full visibility condition if
drivers could see a near point 0.53 s ahead and a distant point
0.93 s ahead.
Steen et al. (2011) reviewed many studies which proposed
one, two or multi-point preview models and concluded that
a two-point preview model was the most realistic, with one
point close to the driver and one more distant point. However, Sharp and Valtetsiotis (2001) used a shift register to
formulate a multi-point preview controller using visual information taken from a single preview point, suggesting that a
human driver in a moving vehicle could use memory to construct a multi-point image of the road geometry from data
sensed at just one or two discrete points. The use of linear
quadratic optimal control theory to calculate the gains on
multi-point road path geometry ahead of the vehicle shows
that the gains eventually tend to zero as the time ahead of
the vehicle increases. This indicates that looking beyond a
certain point might result in diminishing returns (Sharp and
Valtetsiotis 2001; Cole et al. 2006), with the time ahead of
the vehicle at which this occurs dependent on the dynamic
properties of the vehicle and the driver, and the amount of
control effort applied by the driver.
Biol Cybern (2016) 110:91116
et al. (1996) found that perception thresholds for subjects

with vestibular deficiencies were not significantly higher than
for normal subjects, and Bronstein and Hood (1986) found
that neck proprioception largely replaced vestibular function
in vestibular deficient subjects for head rotations relative to
the body. In contrast, Mallery et al. (2010) found that a subject
with vestibular deficiencies had rotational velocity thresholds
an order of magnitude higher than those of normal subjects
and Valko et al. (2012) found that vestibular deficient subjects
had significantly higher perception thresholds in four different motion axes. The relative importance of the vestibular and
somatosensory systems may depend on the precise nature of
the stimuli; however, it does appear that the vestibular system
is an important source of information for drivers.
The vestibular system consists of two sets of organs
located in the inner ear: the semicircular canals (SCCs) which
sense rotational motion and the otoliths which sense translational motion (Kandel et al. 2000). Many studies have
investigated the function of the vestibular system in primates
and humans, either directly by measuring electrical signals
in the brain or indirectly by measuring the vestibulo-ocular
reflex (VOR), a reflexive eye movement which uses vestibular information to compensate for head movements.
2.2.1 Otoliths
The otoliths are formed from small granular particles contained in a gelatinous membrane which is in turn connected to
sensory cells via hairs called cilia. When subjected to translational acceleration, the inertia forces on the otoliths deflect
the cilia and excite the sensory cells (Kandel et al. 2000).
Most mathematical models are based on empirical data from
experiments carried out on humans and animals.
It is a natural extension of Einsteins equivalence principal (Einstein 1907) that humans cannot tell the difference
between a translational acceleration and a change in orientation of the gravity vector. Young and Meiry (1968) developed
a model for the otoliths relating the perceived specific force
(combination of inertial and gravitational accelerations) to
the actual specific force. They proposed the transfer function:

Hoto (s) = K oto
(1 + Toto1 s)
(1 + Toto2 s)(1 + Toto3 s)

(2)
2.2 Vestibular system

There is some disagreement in the literature as to the relative importance of the vestibular system in nonvisual motion
perception. Studies measuring thresholds of human motion
perception in the dark often assume that the influence of the
vestibular system is much larger than that of the somatosensors (Benson et al. 1986, 1989; Grabherr et al. 2008; Soyka
et al. 2012, 2009, 2011; Kingma 2005). However, Gianna
123
and identified values for its parameters, given in the first

row of Table 1. With these values, the transfer function is
essentially low pass but with a constant reduction in gain at
very low frequencies.
Fernandez and Goldberg (1976) measured the afferent firing rate (AFR) in the brains of squirrel monkeys subjected to
accelerations at various frequencies and magnitudes. They
developed a model of the otoliths containing a fractional
Biol Cybern (2016) 110:91116
97
2.2.2 Semicircular canals
Table 1 Otolith model parameters

Study
K oto
Toto1
Toto2
Toto3
Young and Meiry (1968)
0.4
13.2
5.33
0.66
Hosman (1996)
0.4
0.5
0.016
Telban and Cardullo (2005)
0.4
10
0.016
Soyka et al. (2011)
0.0225
22.05
0.62
0.016
Magnitude (abs)
10 0
10 -1
10 -2
The semicircular canals consist of sets of three elliptical cavities which are each filled with fluid (Kandel et al. 2000).
Angular motion about any axis causes the fluid to move
within these cavities, causing deflections of small hair cells
which excite sensory cells. Early models of the SCCs were
based on considerations of the physical dynamics of the
organs. Steinhausen (1933) used observations of the motion
within the SCCs of fish to develop the torsion-pendulum
model. Young and Oman (1969) adapted this model to
include additional adaptation terms TSCCa to match trends
seen in experimental results. Fernandez and Goldberg (1971)
added a lead term TSCC1 , giving the transfer function:
Young and Meiry (1968)

Hosman (1996)
Soyka et al (2011)
HSCC (s) = K SCC
TSCCa s
(1 + TSCCa s)
(1 + TSCC1 s)
(1 + TSCC2 s)(1 + TSCC3 s)
(3)
Phase (deg)
10 -3
90
45
0
-45
-90
10 -3
10 -2
10 -1
10 0
10 1
10 2
10 3
Frequency (rad/s)
Fig. 5 Bode plot for otolith transfer function with parameters from
different studies, given in Table 1. Input is acceleration
exponent, which is difficult to implement practically. Therefore, Hosman (1996) proposed a simplified version in the
same form as Eq. 2. Based on this and other research, Telban and Cardullo (2005) suggested parameters for a transfer
function in the form of Eq. 2, relating the specific force input
to a perceived specific force output. Soyka et al. (2011) used
a signal-in-noise model to find a transfer function for the
otoliths which optimised the fit to sensory threshold measurements (see Sect. 4 for more information). Suggested otolith
parameters from these studies are summarised in the remaining rows of Table 1. The gains K oto have been adjusted to
give comparable outputs, since the scaling of the output signal is arbitrary. Bode plots of the otolith transfer function
using the different parameters are compared in Fig. 5. For
a driving task, the mid-range frequencies (between around
101 and 101 rad/s) are the most important, and in this range
the otoliths exhibit a roughly proportional response to accelerations. There are differences in the details of the frequency
responses measured in different studies, which highlights the
difficulty in achieving repeatable results when using different
subjects, equipment and methodologies.
which relates the AFR to the angular acceleration of the stimulus.

Fernandez and Goldberg (1971) measured the AFR of
squirrel monkeys in response to angular accelerations of various amplitudes and frequencies. Hosman (1996) suggested
alternative parameter values based on results from the literature, neglecting the adaptation time constant TSCCa since it
lies outside the bandwidth of interest for driving tasks. Telban and Cardullo (2005) reviewed several relevant studies
and suggested slight modifications to the parameters of Eq. 3.
They also proposed a simplified transfer function for modelling purposes, which links angular velocity inputs (hence
the s 2 term) to perceived angular velocity outputs:

HSCC (s) = K SCC
TSCCa s 2
(1 + TSCCa s)(1 + TSCC2 s)

(4)
(there is a typographical error in (Telban and Cardullo 2005),

with s in the numerator instead of s 2 ). This transfer function neglects the short time constants TSCC1 and TSCC3 ,
which affect frequencies well above the range of normal head
movements. The key feature of the transfer function is rolloff below about 102 rad/s which means that there is zero
response at constant angular acceleration. In the same way
as for the otoliths (Soyka et al. 2011), Soyka et al. (2012)
chose time constants to optimise the fit to sensory threshold measurements using a signal-in-noise model. Similarly
to Hosman (1996), they neglected the adaptation time constant TSCCa . SCC parameters found from various studies are
summarised in Table 2. As with the otoliths, the gains K SCC
have been adjusted to give comparable outputs. Bode plots
of the SCC transfer function using the different parameters
are compared in Fig. 6. At mid-range frequencies the transfer functions have the characteristics of an integrator, hence
123
98
Biol Cybern (2016) 110:91116
Table 2 SCC model parameters
Study
K SCC
TSCCa
TSCC1
TSCC2
TSCC3
Fernandez and Goldberg (1971)
5.73
80
0.049
5.70
0.005
Hosman (1996)
5.73
(80)
0.110
5.90
0.005
5.73
80
(0.060)
5.73
(0.005)
Soyka et al. (2012)
2.2
()
0.014
2.16
0.005
Parameters which the authors have suggested may be neglected are given in brackets
Magnitude (abs)
10 2
pedals. The following subsections discuss the properties of

the muscle spindles, which measure muscle displacement,
and the Golgi tendon organs, which measure muscle force.
Other somatosensors which may play a role are skin receptors and joint receptors which give information on touch and
joint angle (Collins et al. 2005; Proske and Gandevia 2009),
and graviceptors which respond to the motion of fluid within
the body (Vaitl et al. 2002). While these somatosensors may
give the driver useful information, such as the contact forces
between the body and the seat, the nature of these stimuli
means they are difficult to measure and quantify, and as such
the existing literature does not lend itself to application within
driver models.
10 0
10 -2
Fernandez and Goldberg (1971)

Hosman (1996)
Soyka et al (2012)
Phase (deg)
10 -4
90
45
0
-45
2.3.1 Muscle spindles

-90
10 -4
10 -2
10 0
10 2
10 4
Frequency (rad/s)
Fig. 6 Bode plot for SCC transfer function with parameters from different studies, given in Table 2. Input is angular acceleration
why Telban and Cardullo (2005) suggested the SCCs measure angular velocity rather than acceleration. In contrast to
the otolith dynamics, the agreement between the different
studies is much higher. This may be because these studies
based their work on similar models of the physical dynamics
of the SCCs, although the transfer function found from sensory thresholds (Soyka et al. 2012) also agrees well with the
others at mid-range frequencies.
2.3 Somatosensors
During driving, the information provided by the visual and
vestibular systems is complemented by the response of various receptors of the somatosensory system (Kandel et al.
2000). A particular group of receptors provide proprioception, which is the sensing of joint angles and movements
and muscle displacements and forces. These receptors are
particularly important in allowing the driver to sense the
angle and torque of the steering wheel, which can be used
by experienced drivers to sense the characteristics of the
contact between the tyre and the road. Proprioceptors are
also used to sense the displacements and forces of the foot
123
Muscle spindles are sensors which detect the length and rate
of change of length of the muscles. They produce two separate signals, one dependent on muscle velocity and length
(type Ia afferent) and one dependent on muscle length only
(type II afferent) (Kandel et al. 2000). An empirical linear
model of the muscle spindle response, based on measurements taken in cats, was formulated by Poppele and Bowman
(1970), with the Ia and II afferent responses to muscle displacements given by:
s(s + 0.44)(s + 11.3)(s + 44)
(s + 0.04)(s + 0.816)
(s + 0.44)(s + 11.3)
HII (s) =
(s + 0.816)
HIa (s) =
(5)
(6)
More complicated nonlinear models have also been developed which can predict the afferent responses accurately
under a wide variety of conditions (Maltenfort and Burke
2003; Mileusnic et al. 2006).
2.3.2 Golgi tendon organs
Golgi tendon organs (GTOs) respond to the forces in the
muscles. They share a nerve with the Ia afferent response of
the muscle spindles, giving a response known as a type Ib
afferent (Kandel et al. 2000). A linear model of the GTOs
was first proposed by Houck and Simon (1967), again based
on measurements in cats. Their model was stated as a transfer
Biol Cybern (2016) 110:91116
99
function between muscle force and Ib afferent response by

Prochazka (1999):
HIb (s) = 333
(s + 0.15)(s + 1.5)(s + 16)

(s + 0.2)(s + 2)(s + 37)
(7)
A nonlinear model of the GTOs has also been developed

(Mileusnic and Loeb 2006) and has been found to describe
the static and dynamic properties of the GTOs accurately.
3 Time delays
As shown in Fig. 3, there are various ways in which delays are
introduced between sensory stimuli being applied to a driver
and the drivers control response being measured. Delay
sources include receptor dynamics, nerve conduction, neural
processing and neuromuscular dynamics. Various techniques
have been used in the literature to measure delays in human
response to sensory stimulation. The simplest of these is to
apply a stimulus and measure the time taken for a physical
response (such as pressing a button) to be recorded. Some
studies have used more sophisticated methods of applying stimuli, such as galvanic vestibular stimulation (GVS)
which bypasses the vestibular organs by applying an electrical stimulus directly to the nerves (Fitzpatrick and Day
2004). Other methods have been used to detect responses at
other points in the process, such as measuring the VOR to
identify the reflexive delay, using magnetoencephalography
(MEG, Hmlinen et al. 1993) or electroencephalography
(EEG) to measure electrical impulses within the brain or
using electromyography (EMG) to record electrical activity
in the muscles.
When interpreting sensory time delays measured in different studies using different techniques, it is important to
consider which of the delay components shown in Fig. 3 are
included in the measurement in each case. The aim of this
section is to use results from the literature to estimate the
total delay between stimulus and response for each sensory
system. However, it can be difficult to separate the effects of
pure time delays from lags due to the dynamics of the sensors and muscles and the time taken for signals to rise above
noise levels (Soyka et al. 2013). Nevertheless, results from
the literature can be used to find an approximate estimate
of the order of magnitude of time delays in human sensory
systems.
EMG has been used to measure the response of the muscle spindles to applied muscle stretches, finding delays of
2530 ms for the Ia afferent and 40 ms for the II afferent
(Matthews 1984). Bigler (2013) combined these with measured nerve conduction delays (Trojaborg and Sindrup 1969;
Kandel et al. 2000) to give delays of 34 ms and 48 ms for the
Ia and II afferents. As the Ib afferent response of the GTOs
shares the same nerve as the Ia muscle spindle response, the

time delay for the Ib afferent may be the same as the Ia muscle spindle response. However, these values do not include
any neural processing time, so the actual sensor delays are
likely to be larger.
Reaction times for drivers responses to simulated wind
gusts have been measured in a driving simulator (Wierwille
et al. 1983). Mean delays of 0.56 s without motion feedback
and 0.44 s with motion feedback were found. These measurements encompass the complete process between stimulus
application and physical response shown in Fig. 3, including
all delays, lags and noise. Therefore, they can be considered
as upper bounds for the delays in the visual system and combined visualvestibular systems during driving. MEG has
been used to record neural responses to visual stimuli and
delays of 140190 ms have been found (Kawakami et al.
2002; Lam et al. 2000), although it is unclear how much
neural processing is carried out before and after this response
is measured. Vestibular reflex delays have been measured by
actively stimulating vestibular nerves using GVS and measuring the latency to the onset of the VOR (Aw et al. 2006;
Tabak et al. 1997). Delays of 59 ms have been found, showing that the conduction of vestibular reflex signals is very
fast.
There is a growing body of evidence, reviewed by BarnettCowan (2013) that despite the very fast conduction of
vestibular reflex signals, vestibular processing can take much
longer than the processing of other sensory signals. Vestibular delays have been found to be significantly longer than
visual delays when measuring brain responses using EEG
(Barnett-Cowan et al. 2010) and when measuring overall
reaction times (Barnett-Cowan and Harris 2009). BarnettCowan et al. (2010) measured impulses in the brain 100 ms
and 200 ms after visual and vestibular stimuli, respectively,
with a further 135 ms until a button was pressed in both
cases. This gives visual and vestibular delays of 235 ms and
335 ms; however, Barnett-Cowan (2013) suggested that these
delays may include the time taken for the stimuli to rise above
threshold levels (as modelled by Soyka et al. 2013) so they
may be overestimates.
The visual and vestibular delays measured by BarnettCowan et al. (2010) are significantly lower than those found
in a driving simulator by Wierwille et al. (1983). Furthermore, Barnett-Cowan et al. (2010) measured larger vestibular
delays than visual delays, whereas Wierwille et al. (1983)
found that adding vestibular stimuli significantly reduced
the overall delay. This may indicate that sensory delays are
dependent on the conditions in which the stimuli are applied.
Delays due to nerve conduction and sensory and neuromuscular dynamics are a result of biochemical processes which
are unlikely to depend significantly on the precise nature of
the task carried out. However, it is likely that neural processing time is affected by the complexity of the task and the
123
presence of distracting information and stimuli. Studies have

investigated the intermittent nature of cognitive processing
(Gawthrop et al. 2011; Johns and Cole 2015), which may
play a part in increasing reaction times with increased mental load.
Rather than passively responding to stimuli as in many
of these studies, drivers actively control the motion of the
vehicle. It is difficult to measure time delays during an active
control task, as response times are affected by the closedloop dynamics. Some insight can be gained by looking at
studies which have identified visual and vestibular delays
during closed-loop pilot control tasks (Ellerbroek et al. 2008;
Nieuwenhuizen et al. 2013; Zaal et al. 2009c,a, 2012, 2013).
In general, vestibular delays have been found to be lower than
visual delays, with vestibular delays between around 0.05
0.23 s and visual delays between around 0.180.32 s. These
values seem consistent with the values measured in passive
conditions; however, due to the large variability in measurements it is difficult to say whether delays are longer in active
or passive conditions. Delays have been found to increase in
the presence of additional stimuli (Zaal et al. 2009a) and in
real flight compared with a simulator (Zaal et al. 2012). This
indicates that perceptual delays are higher during multimodal
conditions.
4 Perception thresholds
Due to limits of human sensory organs and noise caused
by spontaneous neuron firing, sensory systems have thresholds below which stimuli cannot be perceived. Perception
thresholds are defined as the smallest stimulus which can
be detected, and these are commonly measured by asking
subjects to distinguish something about the stimulus, such
as its direction. In reality, these thresholds are generally not
precise, but a smooth transition from 0 to 100 % probability
of detection over a range of values. This cumulative probability distribution is known as a psychometric function
(Boring 1917) and is often modelled as a cumulative normal distribution. Variations on the updown method (Levitt
1970) are commonly used to measure perception thresholds,
and depending on the method used the thresholds measured
correspond to different probabilities of detection, generally
between 65 and 80 %.
The just noticeable difference (JND) is defined as the
smallest change in amplitude from a reference stimulus
which is required before the difference between the two stimuli is noticed. From experiments on the perception of lifted
masses, Weber (1834) found that the JND in mass was proportional to the reference mass. This result has been found to
be applicable for many perceptual systems and has become
known as Webers law with the constant of proportionality known as the Weber fraction. Figure 7 shows how the
123
Biol Cybern (2016) 110:91116
JND
100
Weber
Fraction
1
Perception
Threshold
Stimulus Intensity
Fig. 7 JNDs for a stimulus following Webers law. Webers law states
that JNDs increase linearly with stimulus intensity. The constant of
proportionality is known as the Weber fraction, and the y-intercept is
the perception threshold
JND varies with stimulus intensity for a stimulus following

Webers law.
Many of the published measurements of perception
thresholds were carried out under passive, unimodal conditions, meaning that the test subjects were exposed only to
the one stimulus of interest and they did not perform any task
other than perceiving the stimulus. However, during driving
multiple senses are being stimulated simultaneously in different axes, and the driver is carrying out an active control
task. Groen et al. (2006) defined the indifference threshold
as the threshold for perception of a stimulus in the presence
of other congruent or incongruent stimuli. JNDs are a special
case of indifference thresholds, when the background stimulus is in the same axis and modality as the stimulus which
is being detected. Another special case of the indifference
threshold is for congruent stimuli from two different sensory modalities (e.g. visual and vestibular systems), where
the indifference threshold marks out a coherence zone of
stimuli which are perceived as consistent with each other.
4.1 Threshold models
The simplest model of sensory thresholds is a dead zone
where the perceived amplitude is zero. There are two possible
methods for modelling this, as shown in Fig. 8. Method 2
is the most applicable of these, as method 1 implies that
the perceived amplitude would be smaller than the actual
amplitude, even above the perception threshold. The dead
zone model is useful for simplicity; however, it assumes that
the psychometric function is a step function, and it cannot be
used directly to model JNDs.
Recent studies have suggested that sensory thresholds
arise primarily as a result of noise in the sensory channels
and the brain. Soyka et al. (2011, 2012) developed models
of translational and rotational motion perception thresholds
Biol Cybern (2016) 110:91116
101
Method 1
Method 2
Perceived Amplitude
Perceived Amplitude
Stimulus
Amplitude
Stimulus
Amplitude
Fig. 8 Two methods of modelling sensory thresholds as dead zones.

In method 1, the perceived amplitude increases from zero after the
threshold is reached, whereas in method 2 the perceived amplitude is
equal to the stimulus amplitude above the threshold
AN
SDN
Stimulus
Sensor
Transfer
Function
SDN
Gain
+
+
AN
Gain
+
+
are found in this way for the different senses using sensory
threshold measurements from the literature. These measurements have all been taken under passive unimodal conditions;
therefore, since thresholds have been found to increase under
active or multimodal conditions (see Sect. 4.3) the noise
amplitudes found in this section can be considered to be
lower bounds. For each sensory system, the signal-in-noise
model of Soyka et al. (2011, 2012) has been used to identify the additive noise amplitudes using two different transfer
functions: (i) a published sensor transfer function from considerations of the sensory dynamics and (ii) a sensor transfer
function optimised to fit threshold data. It is unclear which
of the two transfer functions is more appropriate for driver
modelling. The parameters derived from sensory threshold
measurements may describe the behaviour at low amplitudes
better; however, they may not completely match the dynamic
behaviour of the sensory system. Noise amplitudes are given
in units with a * symbol at the end, to indicate that the noise is
added to the stimuli filtered by the sensory transfer functions.
4.2.1 Visual thresholds
Perceived
Stimulus
Fig. 9 Sensor model incorporating additive and signal dependent noise

(Bigler 2013). Noise is added after the sensor transfer function to represent spontaneous neuron firing in the brain. This is similar to the model
of Soyka et al. (2011, 2012), who modelled thresholds using a constant
noise addition after the sensory transfer function
based on additive noise (AN) applied to the outputs of the

otolith and SCC transfer functions. The perception thresholds were found as the minimum stimulus amplitude required
for the output to exceed the noise level. Both studies found
good fits to experimental results, although the transfer functions had to be adjusted slightly from those found in the
literature (see Sect. 2). This model predicts the frequency
dependence of perception thresholds and is valid for arbitrary motion inputs rather than solely sinusoidal motion. A
similar principle was used by Bigler (2013) to model JNDs
as well as perception thresholds, by adding signal-dependent
noise (SDN) as well as AN to the output of the sensor transfer function (Todorov 2005). This sensor model is shown in
Fig. 9.
4.2 Passive threshold measurements
Thresholds and JNDs have been measured in passive conditions for a variety of stimuli. Soyka et al. (2011, 2012)
showed that sensory thresholds could be predicted by finding
when the output of the sensory transfer function rises above a
specific noise amplitude; therefore, this model can be used in
reverse to infer noise amplitudes from sensory threshold measurements. In the following subsections, noise amplitudes
Various studies have measured perception thresholds and

JNDs for the visual perception of self-motion. A difficulty in
interpreting these results with any certainty is that they may
well be dependent on the characteristics of the visual scene,
such as the relative motion of stationary reference objects in
the visual field, so it is not clear how generally applicable
the results are. However, it may still be possible to find some
useful information about the performance limits of the visual
system.
A driving simulator display was used by Bigler (2013) to
measure yaw angle and lateral displacement thresholds. The
display was not calibrated to give full-scale visual feedback
so the absolute values of the measured thresholds may not be
at the correct scale; however, the frequency response should
not depend on the display scaling. The results are shown
in Fig. 10. The visual transfer function given in Eq. 1 was
used with the model of Soyka et al. (2011, 2012) to give
predicted thresholds, shown by the solid lines in Fig. 10. The
model fits the thresholds very well, which is not surprising
considering that the visual transfer function was found by
fitting parameters to these results. The additive noise levels
found are 0.0011 rad/s* for the yaw angular velocity and
0.032 m/s* for the lateral velocity.
A fit to the data was also found for a simple model of
the visual system dynamics, with unity transfer functions
between actual and perceived yaw and sway velocities. The
fit using this model is shown by the dotted lines in Fig. 10,
and the noise values found were 0.0013 rad/s* for the yaw
angular velocity and 0.035 m/s* for the sway velocity. Visual
JNDs have been measured for a range of yaw velocities, and
Weber fractions of 7 % (de Bruyn and Orban 1988), 10 %
123
0.01
Measurements
Model (dynamics)
Model (thresholds)
0.008
0.006
0.004
0.002
0
10
10
100
101
Frequency (Hz)
(a)
Yaw
Lateral Displacement Threshold (m)
Biol Cybern (2016) 110:91116

Yaw Angle Threshold (rad)
102
0.25
Measurements
Model (dynamics)
Model (thresholds)
0.2
0.15
0.1
0.05
0
10
10
100
101
Frequency (Hz)
(b)
Sway
Fig. 10 Visual feedback perception thresholds measured by Bigler

(2013). Sinusoidal yaw angle and lateral displacement signals were
superimposed on constant velocity forward motion and the minimum
detectable motion measured. The visual display was a 2-dimensional
screen which was not scaled to match the real-life motion amplitude,
so the absolute values of the thresholds may not be at the correct scale.
Model fits are shown using two different transfer functions, one from
a simple model of the sensory dynamics of the visual system and one
which was optimised to fit the measured thresholds
(dos Santos Buinhas et al. 2013) and 11 % (Nesti et al. 2014b)

have been found. No studies have been found which measure
visual JNDs for lateral motion.
A few studies have investigated the limits of visual perception of motion in the longitudinal direction. Reinterpretation
of the data collected by Bremmer and Lappe (1999) gives a
JND in displacement in the longitudinal direction of 450 mm,
with a reference displacement of 4 m. This gives a Weber
fraction of 10 %; however, extrapolating from measurements
taken for this relatively short displacement of 4 m may
be inaccurate. Monen and Brenner (1994) determined the
smallest step increase in forward velocity necessary for the
difference to be perceived within half a second and found a
large Weber fraction of around 50 %.
Thresholds of visual perception involved in feedforward
control have not been measured explicitly. Authi and Mestre
(2012) measured JNDs in path curvature, finding a Weber
fraction of approximately 11 %. Bigler (2013) used the results
of Legge and Campbell (1981), who found the angular resolution of the retina to be around 1.5 arc min, to calculate
additive and multiplicative noise variances for visual perception of road path geometry ahead of the vehicle. However,
these results were found by asking subjects to indicate when
they could detect a change in position of a small dot, which
is likely to be significantly easier than picking out the full
road geometry from a complicated visual scene.
lar in both directions (Benson et al. 1986); therefore, they are

considered together. Thresholds have also been measured in
the vertical (Z) direction (Nesti et al. 2014a); however, this
is not so relevant for the car drivers control task.
The updown method (Levitt 1970) was used to measure
thresholds in several studies, with participants being subjected to sinusoidal stimuli with amplitudes which changed
for each trial (Benson et al. 1986; Kingma 2005; Soyka
et al. 2009, 2011; Hosman and Van Der Vaart 1978; Heerspink et al. 2005). Other studies used gradually increasing or
decreasing motion amplitudes and asked subjects to indicate
when they started or stopped perceiving motion (Hosman and
Van Der Vaart 1978; Heerspink et al. 2005). The thresholds
for decreasing amplitudes were found to be lower than the
thresholds for increasing amplitudes. It was thought that this
was because the subjects were already tuned in to the signal
so were able to pick it out from the noise more easily. In all
of these studies, the subjects were moved in only one axis at
a time while seated in the dark, so they were focused on the
acceleration stimulus without any other distractions.
Thresholds for the discrimination of the direction of sinusoidal accelerations in the horizontal plane from the studies
using the updown method are shown in Fig. 11. It is clear
that there is a large variability in results between different
studies and even within each study, indicating that perception thresholds are sensitive to differences in experimental
methods and participants.
Predicted thresholds are also shown in Fig. 11, found
using the signal-in-noise model of Soyka et al. (2011). The
transfer function given in Eq. 2 was used with two different sets of parameters from Table 1. The dotted line shows
the results using parameters found by Telban and Cardullo
(2005) from the dynamics of the otoliths and measurements
4.2.2 Otolith thresholds

Perception thresholds have been measured extensively for
translational accelerations in the horizontal plane. Measurements have been carried out in the longitudinal (X) and lateral
(Y) directions, and the thresholds have been seen to be simi-
123
103
0.1
Benson et al. (1986) X
Benson et al. (1986) Y
Soyka et al. (2011) X/Y
Soyka et al. (2009) X/Y
0.08
0.06
0.04
Kingma (2005) X
Kingma (2005) Y
Model (dynamics)
Model (thresholds)
0.02
0
10
10
100
101
Frequency (Hz)
Angular Velocity Perception Threshold (rad/s)
Acceleration perception threshold (m/s2)
Biol Cybern (2016) 110:91116
0.12
Benson et al. (1989) exp. 1
Benson et al. (1989) exp. 2
Grabherr et al. (2008)
Soyka et al. (2012)
Model (dynamics)
Model (thresholds)
0.1
0.08
0.06
0.04
0.02
0
10
10
100
101
Frequency (Hz)
Fig. 11 Lateral (Y) and longitudinal (X) acceleration thresholds measured in several different studies, compared with models found from
the dynamics of the otoliths and from threshold measurements
Fig. 12 Yaw angular velocity thresholds measured in several different

studies, compared with models found from the dynamics of the SCCs
and from threshold measurements
of brain responses, and the solid line shows the results using
parameters optimised by Soyka et al. (2011) to fit the measured thresholds. The threshold model was found using the
results of Soyka et al. (2011) only, whereas the noise level
for the dynamics transfer function was optimised to fit the
whole data set. The noise levels found were 0.038 m/s2 *
for the dynamics transfer function and 0.015 m/s2 * for the
thresholds transfer function. The transfer function found
from the threshold measurements fits the results much better
than the transfer function found from the sensory dynamics,
as the corner frequency at which the thresholds plateau is too
low for the dynamics transfer function.
Naseri and Grant (2012) measured JND values for sinusoidal accelerations at 0.4 and 0.6 Hz with varying amplitudes. The results were found to fit Webers law well,
although a dependence on frequency was also seen. A Weber
fraction of 5 % was found for the measurements taken at
0.4 Hz, whereas a value of 2 % was found for the measurements taken 0.6 Hz.
In interpreting the results of experiments which measure
thresholds of whole body motion, the possibility of multimodal stimuli should be considered. For example, in the case
of sinusoidal angular velocity imposed on the test subject, the
semicircular canals and various somatosensors may be stimulated simultaneously. Multimodal thresholds and sensory
integration are discussed in Sects. 4.3 and 5.
studies using the updown method are shown in Fig. 12. The
data all follow a similar trend, with a fairly low amount of
scatter compared to the otolith results. Predicted thresholds
are also shown using the signal-in-noise model of Soyka et al.
(2012), based on the transfer function given in Eq. 3. The
solid line was found using parameters optimised by Soyka
et al. (2012) to fit the threshold measurements, and the dotted
line was found using the parameters suggested by Telban and
Cardullo (2005) for the SCCs, choosing the noise level to fit
the measured threshold parameters as well as possible. Both
sets of SCC parameters are given in Table 2. The noise levels
found were 0.025 rad/s* for the thresholds transfer function
and 0.023 rad/s* for the dynamics transfer function. Both
models fit the results well, although the model which was
optimised to fit the threshold results matches more closely as
expected.
JNDs for angular velocity perception have been measured
by Mallery et al. (2010) and dos Santos Buinhas et al. (2013),
finding Weber fractions of 3 and 13 %, respectively. The
difference between these values may be a result of the fact
that Mallery et al. (2010) measured JNDs at larger amplitudes
than dos Santos Buinhas et al. (2013). Mallery et al. (2010)
also found that the gradient (JND/amplitude) was higher at
low amplitudes, and suggested a power law should be used
rather than Webers law. However, it is debatable whether
JNDs for the SCCs should follow a power law, when most
other sensory systems have been found to follow Webers
law.
4.2.3 Semicircular canal thresholds

Various studies have measured thresholds for perception of
angular velocity, using either the updown method (Benson
et al. 1989; Grabherr et al. 2008; Soyka et al. 2012) or by
gradually increasing or decreasing amplitudes (Hosman and
Van Der Vaart 1978; Heerspink et al. 2005), in a similar
way to the otolith measurements. Measured thresholds from
4.2.4 Somatosensor thresholds

Various studies have measured perception thresholds for the
displacements of different limbs; however, Bigler (2013) is
thought to be the first to directly measure thresholds for the
123
Average Displacement Threshold (rad)
104
Biol Cybern (2016) 110:91116

0.035
4.3 Active and multimodal thresholds
0.03
0.025
0.02
0.015
0.01
0.005
0
10
10
100
101
10 2
Stimulus Frequency (Hz)
Fig. 13 Thresholds for the perception of steering wheel angular displacement, measured by Bigler (2013). These results could be used to
infer somatosensor noise levels; however, assumptions would have to
be made about the integration of signals from the two muscle spindle
afferents and the GTOs
perception of steering wheel angle, finding the results shown

in Fig. 13. These results cannot be used to find noise levels for the somatosensors without making some assumptions
about the relationship between steering wheel displacement
and the displacements, velocities and forces of the muscles,
and further assumptions about the method used to integrate
information from the Ia, Ib and II afferents. Further work
is necessary to determine appropriate noise levels for the
somatosensors.
Newberry et al. (2007) measured JNDs in steering wheel
angle and reported a Weber fraction of 14 %. However, this
was achieved by fitting a line with zero perception threshold, and a better fit to the data can be achieved by including
the effect of a nonzero perception threshold. This gives a
good linear fit to the measurements, with a Weber fraction
of 9.6 % and a perception threshold of 0.006 rad. The stimulus profile and frequency were not reported by Newberry
et al. (2007); however, the extrapolated perception threshold
is similar to that measured by Bigler (2013) for stimuli at
1 Hz (see Fig. 13).
To date, no studies appear to have directly measured
perception thresholds for steering wheel force or torque.
Steering wheel force JNDs were measured by Newberry et al.
(2007), and extrapolating from these measurements gives a
perception threshold of 0.45 N and a Weber fraction of 9.6 %.
It is interesting to note that the Weber fraction found for the
GTOs using this method matches the Weber fraction found
for the muscle spindles almost exactly, suggesting that there
may be a perceptual link between the two sensors. The GTO
afferent and the primary muscle spindle afferent share the
same nerve conduction path (Kandel et al. 2000), so since
JNDs are related to noise along the transmission path this
may provide an explanation for the similarity in Weber fraction values.
123
The studies summarised in Sect. 4.2 were all targeted at measuring thresholds of a single stimulus in isolation, during
passive conditions where the subject was concentrating on
the stimulus. However, sensory stimuli which occur during
driving are very different to the stimuli applied in these controlled studies, so these results may not be directly applicable
to driving tasks. Stimuli in driving tasks are perceived under
active rather than passive conditions, and there are several
stimuli being perceived at once. Therefore, the indifference
threshold [threshold in the presence of other stimuli (Groen
et al. 2006)] should determine the limits of perception during
driving.
By asking subjects to perform a secondary control task
in a separate motion axis, it has been found that increasing
the mental load on subjects causes an increase in perception thresholds (Hosman and Van Der Vaart 1978; Samji and
Reid 1992). It should be noted that in both of these studies the
subjects were still actively concentrating on perceiving the
motion cues as well as completing the secondary task. Due
to the equivalence of translational accelerations and shifts in
the gravity vector, the brain can easily be fooled into misinterpreting the two types of motion. Groen and Bles (2004)
and Pretto et al. (2014) found that presenting subjects with
visual cues simulating a translational acceleration while they
were undergoing rotational motion caused threshold of perception of the rotation to increase by factors of 56. Pretto
et al. (2014) also measured thresholds during an active control task and found that they increased by factors up to 4 for
some subjects, but didnt change at all for others. The participants whose thresholds did not increase during the active
driving task reported higher levels of immersion in the simulation, indicating that the sense of realism of the simulation
was linked to participants ability to perceive the motion cues
accurately.
Pitch and roll thresholds have been measured with masking vertical motion cues, finding a significant linear increase
in pitch and roll thresholds with vertical amplitude (Zaichik
et al. 1999; Rodchenko et al. 2000). In contrast to these studies, Valente et al. (2006) found no significant effect of vertical
motion amplitude on pitch rate thresholds. In this study, the
pitch and vertical motion were applied at the same frequency,
which may have caused the motion cues to be perceived as
coherent, making it easier to detect the pitch cues.
Groen et al. (2006) analysed the data of Groen and Bles
(2004) and showed that indifference thresholds for pitch rotation in the presence of visual longitudinal cues follow the
same frequency response as the perception thresholds measured in passive conditions, but are increased by a constant
gain. They used this result to hypothesise that the presence of
additional sensory stimuli scales perception thresholds by a
constant gain, without affecting the frequency response. This
Biol Cybern (2016) 110:91116
105
is consistent with the models of Soyka et al. (2011, 2012) and

Bigler (2013) (shown in Fig. 9), where the threshold is placed
after the sensory transfer function and the additional stimuli
cause an increase in the noise level. Groen et al. (2006) suggested that the increase in noise level is linearly dependent on
the amplitude of the additional stimulus, which is equivalent
to Webers Law in the special case of the additional stimulus being in the same axis and modality as the measured
stimulus.
Recent studies have used parameter identification methods
to estimate threshold values during an active control task in
the same axis, and thresholds in active conditions have been
found to be around 1.6 times larger than thresholds measured
in passive conditions (Pool et al. 2012; Valente Pais et al.
2012).
It is evident from the literature that various factors can
cause thresholds to increase from values measured in passive
conditions, including mental load, the presence of other stimuli and carrying out an active control task. It may therefore
not be appropriate to rely on passive threshold measurements
to model sensory dynamics during an active driving task.
4.4 Coherence zones
The term coherence zone was coined by van der Steen
(1998) to describe the range of amplitudes of inputs to
two sensory systems (such as visual and vestibular systems)
which are perceived as consistent with each other, as shown
in Fig. 14. The coherence zone can be defined in terms of
Vestibular Stimulus Amplitude
GMC = Vestibular Amp.

Visual Amp.
one-to-one
line
PMC
CZW
Coherence
Zone
Visual Stimulus Amplitude
Fig. 14 Coherence zone between visual and vestibular stimuli. For a

given visual stimulus, there will be an upper and lower limit of vestibular stimulus amplitude which is perceived as coherent with the visual
stimulus. The coherence zone width (CZW) is the difference between
these two limits, and the point of mean coherence (PMC) is defined
as the point halfway between the limits. The gain of mean coherence
(GMC) is defined as the ratio of the vestibular amplitude to the visual
amplitude at the PMC and represents the preferred gain between the
visual and vestibular cues
the point of mean coherence (PMC), coherence zone width

(CZW) and gain of mean coherence (GMC) as shown.
Coherence zones between the visual and vestibular systems have been measured at various amplitudes and frequencies (van der Steen 1998; Valente Pais et al. 2010a,b).
The GMC was found to decrease with increasing stimulus
amplitude, with subjects preferring larger vestibular motion
than visual motion at low amplitudes and the opposite being
seen at larger amplitudes. Significant differences were found
between the values measured in different studies, highlighting the fact that coherence zones are highly dependent on
the experimental conditions. Contrary to the results found
for perception thresholds, coherence zones were found not
to change significantly during an active control task (Valente
et al. 2011). This indicates that the perceptual mechanisms
behind perception thresholds and coherence zones may not be
directly linked, and suggests that coherence zones measured
in passive conditions may be applied to conditions where
active control tasks are being carried out.
The concept of a coherence zone has been extended to the
detection of heading direction (de Winkel et al. 2010) and
phase differences (Grant and Lee 2007; Jonik et al. 2011).
Jonik et al. (2011) found that inertial motion can lead visual
motion by up to 22 without the difference being detected.
This result was independent of the stimulus frequency, showing that humans can be considered as phase-error detectors
rather than time delay detectors.
Research has shown that, when asked to tune inertial
motion to match visual motion, subjects pick higher amplitudes when tuning downwards from high amplitude motion
than when tuning upwards from low amplitude motion (Correia et al. 2010). Correia Grcio et al. (2013) defined the
optimal zone as the area between these upper and lower
optima and found that it lay within the coherence zone.
Similar to the PMC, GMC and CZW for coherence zones,
the optimal zone was defined in terms of the point of
mean optimal gain (PMO), gain of mean optimal (GMO)
and optimal zone width (OZW). The GMO was found to
decrease at higher amplitudes and at higher frequencies. In
contrast to coherence zone measurements, the OZW was
found not to vary with amplitude or frequency. By varying
the field of view, resolution and depth of the visual scene,
Correia Grcio et al. (2014) found that the optimal gain is
strongly affected by the quality of the visual cues, with
more realistic visual scenes giving GMOs closer to 1.
Two approaches to modelling CZWs were compared by
dos Santos Buinhas et al. (2013), one matching the perceived
intensity of the two stimuli and applying this to averaged
JNDs, and one summing the JNDs for the two individual
stimuli. Comparison of model predictions with experimental data showed that summing JNDs provides the best fit to
the measured data, explaining the results particularly well at
lower amplitudes. dos Santos Buinhas et al. (2013) suggested
123
106
Biol Cybern (2016) 110:91116
that PMCs could be modelled using Stevens power functions

of perceived stimulus intensity; however, this method was not
experimentally verified.
5 Sensory integration
The sensory systems described in Sect. 2 provide the central nervous system (CNS) with measurements (or sensory
cues) which can be used to estimate vehicle states while
driving. However, these measurements are shaped by the sensor dynamics and also contain additive and signal-dependent
noise (as described in Sect. 4). The CNS must therefore carry
out sensory integration to give a single estimate of the vehicle states from the noisy, filtered information received from
each of the sensors.
In a real-world driving scenario, the driver will be presented with coherent sensory information. Any discrepancies
between information from the different sensors are due
to sensory noise, or incomplete information available to a
particular sensor. However, in some situations the information presented to the different senses may be incoherent or
biased, in which case the driver may use a different integration strategy. This is particularly relevant for motion
in virtual environments, where the visual, vestibular and
somatosensory information presented to the driver may not
all accurately reflect the real-world stimuli. An overview
of methods and results from investigations of sensory integration in a variety of virtual environments (not specific to
driving) is given by Campos and Blthoff (2012). The following subsections build on this, focusing in more depth on
results which suggest how information from the sensory systems summarised in Sect. 2 may be integrated during driving.
5.1 Integration of coherent sensory measurements
The simplest model of sensory integration is a linear weighting of the estimates from different sensory systems (Hosman
and Stassen 1999). Appropriate weightings can be found
using sensory experiments; however, the scope of models
with fixed weightings is likely to be limited. For many
sensory systems, the CNS has been found to integrate measurements using statistically optimal methods (Ernst and
Banks 2002; Oru et al. 2003; Butler et al. 2010; Seilheimer
et al. 2014). These methods are based on Bayes theorem
(Bayes 1763), which relates the a posteriori probability
P( I| S) of condition I given observation S to the probability
P( I| S) of observation S given condition I, the a priori probability P( I), and the observation probability P( S) (which is
usually assumed uniform):
P( I| S) =
123
P( S| I)P( I)
P( S)
(8)
Optimal integration of sensory cues involves choosing

from the set of all possible conditions I={ Ii |i=1, . . . , N I }
the condition Ii which has the highest probability P( Ii | S)
based on the set of observations S = { Si |i = 1, . . . , N S }
from the different sensory channels. For a continuous set
of possible conditions I a probability density function of
P( I| S) can be plotted. Equation 8 shows that P( Ii | S)
depends on an assumption about the probability distribution
P( I) before the measurements are made, known as a prior.
There are various ways in which the optimal value of I can
be chosen, such as the maximum a posteriori (MAP) estimate, the minimum mean square error estimate (MMSE)
and the maximum likelihood estimate (MLE) (Clark and
Yuille 1990; Vaseghi 2005). However, if the priors P( I) and
P( S) are uniform and the probability distributions are symmetric, these estimates will be identical and can found by
maximising the likelihood function P( S| I).
If the probability distributions of the sensory estimates
Si are all Gaussian, the MLE S of a property is found by
weighting each estimate in proportion to the inverse of its
variance i2 (Yuille and Blthoff 1996):
S =

i
1/ 2
wi Si with wi = i 2
j 1/ j
(9)
The variance 2 of the combined estimate S is found from

Eq. 10 to be lower than the variances of the individual estimates from the different sensory systems:

1
1
2 =
i2
i
(10)
Oru et al. (2003) showed that a Gaussian prior can be

included in the MLE analysis as an additional input, weighted
by the inverse of its variance as usual. MacNeilage et al.
(2007) used this result to model the integration of visual and
inertial cues to disambiguate between an acceleration and a
shift in the gravity vector, incorporating priors to model the
assumptions that humans are normally in an upright position
and that smaller accelerations are more likely than larger
ones. Soyka et al. (2015) measured off-centre yaw rotation thresholds and found that SCC and otolith signals were
integrated, although the results suggested information from
additional sensory systems may also have been used.
Near-optimal Bayesian integration of visual and vestibular
information has been measured in several studies (Gu et al.
2008; Butler et al. 2010; Fetsch et al. 2009; Prsa et al. 2012;
Drugowitsch et al. 2014; Fetsch et al. 2010). In contrast to
these studies, de Winkel et al. (2013) only found results that
fit the MLE model for 3 out of 8 participants and Nesti et al.
(2015) found that combined visual-inertial thresholds were
higher than predicted by a MLE model. Butler et al. (2011)
Biol Cybern (2016) 110:91116
107
found that participants exhibited optimal visualvestibular

integration 90 % of the time with a stereoscopic visual display
compared with 60 % of the time with a binocular display.
This suggests that the realism of the visual scene may affect
whether or not visual and vestibular information is integrated
optimally. Some studies have found a slight over-weighting
of one sense with respect to the other, although some of these
have found that vestibular cues are weighted higher (Fetsch
et al. 2009; Butler et al. 2010) while others have found that
visual cues are weighted higher (Prsa et al. 2012). Prsa et al.
(2012) suggested that over-weighting of otolith signals and
under-weighting of SCC signals may occur when vestibular
cues are integrated with visual cues.
In order to develop effective and efficient control strategies for interacting with their surroundings, humans use their
experience to develop internal models of themselves and the
world around them (Wolpert and Ghahramani 2000). They
are able to use learning methods to adapt these models to
changes in the environment (Wolpert et al. 2011) such as
astronauts entering microgravity (Carriot et al. 2015). Using
an internal model, a recursive state estimator can be used
to provide new a priori estimates at each time step to give
improved estimates of the system states. A common implementation of this method is the Kalman filter (Kalman 1960;
Grewal and Andrews 2001). It is assumed that the observer
has an internal model of the system given in state-space
form:
x(k + 1) = Ax(k) + Bu(k) + w(k)
y(k) = Cx(k) + v(k)
(11)
The main difference between a driver and a passenger is

that the driver has perfect knowledge of the inputs u; however, they are perturbed by process noise w. Both driver and
passenger measure the outputs y, which are perturbed by
measurement noise v. The new estimate of the states x(k + 1)
is predicted by propagating the current input u(k) and state
estimate x(k) through the internal model of the system. A correction is then added based on the error between the previous
estimated output Cx(k) and measured output y(k), weighted
by the Kalman gain K(k):

x(k + 1) = Ax(k) + Bu(k) + K(k) y(k) Cx(k)
have also been used to model estimation of vehicle states for

pilots (Onur 2014) and drivers (Bigler 2013).
One implication of MLE models of human sensory integration is that the observer must have access to estimates
of the noise variance for each sensory channel. Ernst and
Blthoff (2004) suggested that the variance may be determined by looking at the responses over a population of
independent neurons. Several studies have attempted to build
realistic neural models to describe this behaviour (Deneve
et al. 1999; Pouget et al. 2000; Barber et al. 2003), and
they have found that a close approximation to MLE can
be achieved in such a way. Fetsch et al. (2009) studied the
integration of visual and vestibular cues to heading angle in
humans and monkeys with varying reliability of the visual
cues. They found that both humans and monkeys were able
to dynamically re-weight the cues between trials, indicating
that they were able to obtain a measure of the reliability of
each cue.
5.2 Integration of biased sensory measurements
While MLE is an optimal method of combining measurements from noisy sensory channels with the same mean, if
the signals are biased such that their means are no longer
coherent, using MLE will cause the bias to carry through
into the optimal sensory estimate as seen in Fig. 15 (Ernst
and Luca 2011). There will always be differences between
the measurements Si from different noisy sensory channels;
however, it is impossible to separate these differences into
those which occur because of stochastic variations about
the mean and those which are a result of biases in the sensory channels without prior knowledge of the values of these
biases. It has been found that the CNS may ignore the discrepancies and integrate the biased sensory measurements using
the MLE method if conflicts are small (Scarfe and Hibbard
(12)
The time-varying Kalman gain K(k) is calculated to give

a statistically optimal estimate using MLE, weighting the
estimates based on the covariances of the Gaussian noise
w and v. If the covariances are time invariant, a steady-state
linear filter can be found to give the optimal state estimates for
the system. Various studies have proposed models of visual
vestibular integration based on Kalman filters (Borah et al.
1988; Zupan et al. 2002; Young 2011), and Kalman filters
Fig. 15 Maximum likelihood integration of two biased sensory channels. The probability distributions of the sensors S1 and S2 (shown by
the solid lines) are both biased, i.e. their means do not correspond to
the true value of I. Using MLE causes this bias to carry through to the
combined estimate given by the dashed line. Also note that the sensor
with the largest variance does not necessary have the largest bias
123
108
Biol Cybern (2016) 110:91116
2011; Butler et al. 2010, 2014) or if the conflicting information is presented in different motion axes (Kaliuzhna et al.
2015). de Winkel et al. (2015) found that over half of subjects
integrated visual and inertial heading information regardless
of the size of the bias. However, other studies have found
evidence of various strategies for reducing bias in perceived
signals (Krding et al. 2007; Landy et al. 1995; Burge et al.
2010; Zaidel et al. 2011).
When presented with two different sensory cues, the CNS
must decide whether or not they are coherent (originating
from the same source). If they are coherent, the difference
between them can be assumed to be a result of stochastic
variations and the cues can be combined using MLE. If not,
the cues should be treated separately, treating the situation as
a cue conflict. Krding et al. (2007) proposed a model using
Bayes rule to decide whether or not two cues are coherent
based on a prior describing the likelihood of the cues coming from the same source. They validated the model using
experimental results; however, Seilheimer et al. (2014) noted
that Krding et al. (2007) did not vary the reliability of the
cues, so it is still uncertain whether their model is valid in
all cases. A similar Bayesian model incorporating priors was
proposed by Knill (2007). They found that the weight applied
to a cue shrunk as the size of the conflict increased, but it
did not decrease to zero. However, other studies have found
that under some circumstances humans will veto a cue that
does not fit with the other sensory measurements (Girshick
and Banks 2009; Landy et al. 1995).
Ghahramani et al. (1997) proposed an additional stage of
cue calibration before cues are fully integrated, where the
difference between the estimates is reduced. The values of
estimates S1 and S2 are calibrated by adding S1 and S2 ,
given by:
S1 = C1 ( S2 S1 )
S2 = C2 ( S1 S2 )
(13)
This improves internal consistency (Burge et al. 2010),

ensuring that the estimates from different sensory systems
agree with each other, although it does not necessarily
improve external accuracy (the overall accuracy of the
CNSs combined estimate). If C1 + C2 = 1, calibration can
achieve full internal consistency by adjusting both estimates
to the same value; otherwise, a smaller reduction in the difference between the estimates is found.
Several studies have shown that vision dominates the other
senses under certain conditions (Rock and Victor 1964; Ernst
and Banks 2002), so a model of visual capture has been
proposed where vision completely dominates the combined
estimate (Ernst and Banks 2002; Ernst and Blthoff 2004). In
such a case, the visual estimate does not change and the other
estimate adapts to be the same as the visual estimate, giving
123
Ci = 0 for the visual channel and Ci = 1 for the other channel. Alternatively, Ghahramani et al. (1997) proposed that
the calibration stage follows a similar weighting structure
to the integration process, based on reliability. He hypothesised that each calibration constant C1 is proportional to the
variance i2 of the sensory estimate Si . Burge et al. (2010)
tested this model in an experiment on visual-haptic estimation of slant and reported strong evidence in favour of this
reliability-based calibration.
Reliability-based calibration does not make physical sense
as a method for reducing sensory bias, however, as the reliability of a cue is independent of its bias (Ernst and Luca 2011).
A sensory estimate could have low variance (high reliability)
and high bias, or a high variance (low reliability) but low bias.
For example, in Fig. 15, the cue with the higher variance has
the lower bias. Testing a cue veto model of integration of
biased sensory estimates, Girshick and Banks (2009) found
that the vetoed cue was not necessarily the cue with the highest variance. Zaidel et al. (2011) compared reliability-based
calibration with fixed-ratio calibration, where the calibration
constants Ci were assumed to be learned from past experience. Fixed-ratio calibration fitted their results better than
reliability-based calibration, with higher weighting placed on
the visual estimate. The sum of the calibration constants C1
and C2 was found to be less than 1, so full internal consistency was not achieved. Zaidel et al. (2011) also explained
how the presence of fixed-ratio calibration could cause erroneous indications of reliability-based calibration to appear
using the methods of Burge et al. (2010). It therefore seems
that with biased sensory information humans may use fixed
calibration constants based on past experience, rather than
change the weightings based on cue reliability.
Linear cue calibration for visualvestibular integration has
been observed in several studies although, as with coherent measurements, there is disagreement about which sense
is more highly weighted. Visual dominance was found by
Rader et al. (2011), whereas the vestibular system was found
to dominate by Harris et al. (2000). Ohmi (1996) found that
visual cues dominated when conflicts were small, but vestibular cues dominated when conflicts were large. Zacharias and
Young (1981) found that vestibular cues dominated visual
cues at higher frequencies, so the dominant sensory system
may depend on the frequency content of the tasks carried
out.
Experimental studies have shown that when a consistent
conflict is observed between the visual and vestibular systems, the perceived motion will eventually drift towards the
visual estimate (Ishida et al. 2008). van der Steen (1998) proposed a model of the optokinetic influence, where the visual
estimate attracts the vestibular estimate over a transient
period, modelling the onset of visual self-motion (vection).
This is modelled by passing the difference between the visual
and vestibular estimates through a low-pass filter given by:
Biol Cybern (2016) 110:91116
HOK (s) =
109
1
1 + OK s
(14)
giving the optokinetic influence which is then added to the

vestibular output as shown in Fig. 16. An implication of this
model is that pre-filtering the vestibular cues by the inverse
of the vestibular dynamics, or conversely pre-filtering the
visual cues by the vestibular dynamics, should cause the
visual and vestibular cues to be perceived as coherent even
though they differ substantially. Wentink et al. (2009) tested
this hypothesis using subjective feedback from experiments
in a simulator and found that pre-filtering the vestibular cues
by the inverse of the vestibular dynamics did indeed result in
coherent perception. However, pre-filtering the visual cues
by the vestibular dynamics produced cues which were perceived as coherent for only half the motion conditions.
Zacharias (1977) and Zacharias and Young (1981) developed a detailed empirical model of visualvestibular integra-
Fig. 16 Model of the optokinetic influence proposed by van der Steen

(1998). The difference between the visual and vestibular estimates is
passed through a low-pass filter HOK (s) and then added to the vestibular
estimate
Vestibular
cue
tion under cue-conflict conditions, shown in Fig. 17. Borah

et al. (1988) developed an adaptive version of the model of
Zacharias (1977), using a slightly modified weighting function then multiplying the visual estimate by the gain K and
combining it with the vestibular estimate using a Kalman
filter. Telban and Cardullo (2005) adapted the model of
Zacharias (1977), using some of the modifications suggested
by Borah et al. (1988) and including the optokinetic influence
modelled by van der Steen (1998). They ran simulations to
find the response to velocity step inputs to the visual system
with and without corresponding vestibular cues and found
that it was possible to reproduce latencies measured in previous studies on humans. However, further validation work
is needed to determine whether this model is more generally
applicable.
Wright et al. (2005) subjected participants to vertical
motion with conflicting visual and vestibular information,
playing back recordings of the visual surroundings of the
apparatus to give a realistic scene. Visual and vestibular cues
were presented with different amplitudes and in some cases
out of phase with one another. The results were found to be
incompatible with linear weighting conflict models and the
more complicated model of Zacharias (1977), as for high
visual amplitudes the visual perception was found to dominate, independent of the vestibular amplitude. More research
is clearly needed to develop a model which can describe
integration of biased sensory estimates under a wide range
of conditions.
Vestibular
Dynamics
0.5
+
HP Filter
Vestibular
Dynamics
(Internal Model)
Visual
cue
0.5+
+
+
Perceived
motion
Visual
Dynamics
Fig. 17 Visualvestibular integration model proposed by Zacharias

(1977). Figure adapted from Telban and Cardullo (2005). The visually
perceived motion is filtered through an internal model of the vestibular
dynamics, producing an expected vestibular signal which is subtracted
from the actual vestibular signal to give an error term. In order to allow
resolution of steady-state cue conflicts, the error is washed out using a
high-pass filter. The weighting on the visual and vestibular cues depends
on the gain K , which varies between 0 and 1 based on symmetric

weighting functions, chosen as cosine bell functions for simplicity. A
large conflict will drive K to zero, vetoing the visual cue [matching the
vestibular dominance seen for large conflicts (Ohmi 1996)]. A small
conflict will cause K to be between 0.5 and 1, depending on the magnitude of the visual cue. This behaviour was chosen to qualitatively match
the transient perception of motion observed in cue-conflict situations
123
110
Biol Cybern (2016) 110:91116
Table 3 Summary of sensory system results from the literature

System
Input
From sensory dynamics
From perception thresholds
Weber
fraction (%)
Sensor
delay (ms)
0.0011 (rad/s*)
711
100560
0.032 (m/s*)
711
100560
1050
100560
11
100560
0.015 (m/s2 *)
25
5440
0.025 (rad/s*)
313
5440
Transfer function
Noise
Transfer function
Noise
0.810
s + 0.810
Visual feedback
Yaw angular
velocity
0.0013 (rad/s*)
Visual feedback
Lateral velocity
0.035 (m/s*)
Visual feedback
Longitudinal
velocity
0.810
s + 0.810
Visual
feedforward
Target path
Preview model
Otoliths
Acceleration
SCCs
Angular velocity
Muscle spindles
(Type Ia)
Arm muscle
displacement
0.4(1 + 10s)
(1 + 5s)(1 + 0.016s)
5.73(80s 2 )
(1 + 80s)(1 + 5.73s)
s(s + 0.44)(s + 11.3)(s + 44)
(s + 0.04)(s + 0.816)
Muscle spindles
(Type II)
Arm muscle
displacement
GTOs
Arm muscle force
0.038 (m/s2 *)
0.023 (rad/s*)
0.0225(1 + 22.05s)
(1 + 0.62s)(1 + 0.016s)
2.2s(1 + 0.014s)
(1 + 2.16s)(1 + 0.005s)
10
>34
(s + 0.44)(s + 11.3)
(s + 0.816)
10
>48
333(s + 0.15)(s + 1.5)(s + 16)

(s + 0.2)(s + 2)(s + 37)
10
>34
For the key sensory systems involved in driving, transfer functions between the input stimulus and the sensory response are given, either from
considerations of the sensory dynamics or from perception threshold measurements. Noise levels have been calculated from sensory threshold
measurements, as well as Weber fractions showing how thresholds increase with stimulus amplitude. Estimates of sensory delays are also included
6 Discussion
Key results from the literature on human sensory dynamics
have been presented in Sects. 2 to 5. In this section, these
results are summarised and discussed with a view to understanding and modelling driver steering and speed control.
Results for the human sensory systems which are most relevant to driver modelling are summarised in Table 3. Transfer
functions are presented which have either been found from
models of the sensory dynamics and measurements of brain
activity or inferred from sensory threshold measurements.
Using the transfer functions found from sensory threshold
data may give more accurate results near the limits of perception; however, they may not capture all of the dynamic
behaviour of the sensory system.
Noise magnitudes have been inferred from sensory threshold measurements using the signal-in-noise model of Soyka
et al. (2011, 2012). These were found from passive threshold measurements taken for one sensory stimulus at a time;
however, thresholds have been found to increase in active
conditions and in the presence of other sensory stimuli by
factors between 1.5 and 6 (Hosman and Van Der Vaart 1978;
Samji and Reid 1992; Zaichik et al. 1999; Rodchenko et al.
2000; Groen and Bles 2004; Valente Pais et al. 2012). This
means that the noise values shown in Table 3 should be considered as lower bounds. Most sensory systems have been
found to approximate Webers law, with JNDs increasing
with stimulus amplitude; therefore, Weber fractions have
123
been included in Table 3. This increase in sensory noise

with stimulus amplitude can be modelled by including signal dependent as well as additive noise (Todorov 2005; Bigler
2013).
Estimates of sensor delays are also given in Table 3 for
each system, comprising of all components of the time delay
between stimulus application and physical response. However, there is still some uncertainty about the precise values,
as it is thought that delays in neural processing may be dependent on the exact nature of the stimuli and the task being
carried out. It is unclear whether delays increase or decrease
during active conditions; however, they have been found to
increase with additional stimuli in multimodal conditions.
For many types of stimuli, coherent sensory information
has been found to be integrated in a statistically optimal fashion (Ernst and Banks 2002; Oru et al. 2003; Butler et al.
2010, 2011; Seilheimer et al. 2014; Gu et al. 2008; Prsa et al.
2012; Drugowitsch et al. 2014). Humans build up internal
models of themselves and their surroundings (Wolpert and
Ghahramani 2000) and a Kalman filter can be used to model
optimal sensory integration using internal models (Kalman
1960; Grewal and Andrews 2001; Borah et al. 1988; Zupan
et al. 2002; Young 2011; Onur 2014). For incoherent sensory information, when conflicting information is presented
to the different sensory channels, sensory integration is less
well understood. A variety of different models have been proposed; however, no overwhelming evidence has been found
in favour of any of them. Consideration of how humans inte-
Biol Cybern (2016) 110:91116
grate incoherent or biased sensory measurements may be

important when studying drivers in virtual environments;
however, in normal driving the senses should be in agreement.
Since sensory parameters have been found to change under
active or multimodal conditions, it may not be appropriate to
apply the results shown in Table 3 directly to a driver model.
It is very difficult to measure sensory parameters directly
during realistic driving conditions. However, by developing
a model of driver control behaviour incorporating sensory
dynamics, parametric identification methods could be used
to gain some insight into the performance of sensory systems
while driving. Parametric identification procedures have been
described by Ljung (1999) and Zaal et al. (2009b) and applied
to driver models by Keen and Cole (2012) and Odhams and
Cole (2014). A number of studies have been carried out
at Delft University of Technology to identify pilot control
strategies under different conditions (Pool et al. 2008; Zaal
et al. 2013, 2009c, 2010, 2009a, 2012; Nieuwenhuizen et al.
2013; Ellerbroek et al. 2008; Drop et al. 2013). Since the
pilot control task is similar to a driver steering control task,
the results of Zaal et al. (2009c) have been used to validate
an initial concept for a driver model incorporating sensory
dynamics (Nash and Cole 2015).
It is hoped that the information presented in this literature review will inform and motivate future researchers to
consider the influence of sensory dynamics in driving tasks.
The differences highlighted here between active and passive
measurements mean that we do not advocate direct incorporation of results from the sensory perception literature into
driver models. Rather, identification methods used in recent
aerospace studies seem to show more promise, and there
is clear scope for applying similar techniques to studying
drivers. The danger in such an approach is that identification of a large number of sensory parameters may become
infeasible. Therefore, care must be taken to increase the complexity of sensory models slowly and use carefully designed
experiments to isolate different features of sensory perception during driving. This review should serve as a guide for
potential areas of investigation and a reference to compare
with new results.
7 Conclusion
The results summarised in this literature review give an
insight into various different sensory systems, and how they
can be used to model driver control behaviour. Sensory transfer functions have been studied extensively, and there is little
disagreement between different studies. Sensory integration
is reasonably well understood under normal conditions; however, there is little agreement on how humans cope with
conflicting sensory information. Studies have shown that
111
sensory thresholds increase under active and multimodal conditions, but further research is necessary to determine how
and why this happens. Time delays also increase during multimodal conditions; however, it is not clear whether they vary
during active control tasks. There is a great deal of scope
for improvement in the available knowledge on human sensory perception during active control tasks, so future research
should focus in this area. It is hoped that the information in
this review will prove useful in developing more sophisticated driver steering and speed control models which take
account of the drivers sensory dynamics.
Acknowledgments The authors wish to thank the anonymous reviewer
who provided many useful suggestions for improving the paper, particularly regarding visual perception.
Open Access This article is distributed under the terms of the Creative
Commons Attribution 4.0 International License (http://creativecomm
ons.org/licenses/by/4.0/), which permits unrestricted use, distribution,
and reproduction in any medium, provided you give appropriate credit
to the original author(s) and the source, provide a link to the Creative
Commons license, and indicate if changes were made.
References
Abbink Da, Mulder M, Van Der Helm FCT, Mulder M, Boer ER
(2011) Measuring neuromuscular control dynamics during car following with continuous haptic feedback. IEEE Trans Syst Man
Cybern Part B Cybern 41(5):12391249. doi:10.1109/TSMCB.
2011.2120606
Authi CN, Mestre DR (2012) Path curvature discrimination: dependence on gaze direction and optical flow speed. PloS One
7(2):e31,479. doi:10.1371/journal.pone.0031479
Aw ST, Todd MJ, Halmagyi GM (2006) Latency and initiation of the
human vestibuloocular reflex to pulsed galvanic stimulation. J
Neurophysiol 96(2):925930. doi:10.1152/jn.01250.2005
Barber MJ, Clark JW, Anderson CH (2003) Neural representation
of probabilistic information. Neural comput 15(8):18431864.
doi:10.1162/08997660360675062
Barnett-Cowan M, Nolan H, Butler JS, Foxe JJ, Reilly RB, Blthoff
HH (2010) Reaction time and event-related potentials to visual,
auditory and vestibular stimuli. J Vis 10(7):e1400. doi:10.1167/
10.7.1400
Barnett-Cowan M (2013) Vestibular perception is slow: a review. Multisens Res 26(4):387403. doi:10.1163/22134808-00002421
Barnett-Cowan M, Harris LR (2009) Perceived timing of vestibular
stimulation relative to touch, light and sound. Exp Brain Res
198(23):221231. doi:10.1007/s00221-009-1779-4
Bayes T (1763) An essay towards solving a problem in the doctrine of
chances. Philos Trans R Soc Lond 53:370418. doi:10.1098/rstl.
1763.0053
Benson AJ, Spencer MB, Stott JR (1986) Thresholds for the detection
of the direction of whole-body, linear movement in the horizontal
plane. Aviat Space Environ Med 57:10881096
Benson AJ, Hutt EC, Brown SF (1989) Thresholds for the perception of
whole body angular movement about a vertical axis. Aviat Space
Environ Med 60:205213
Bigler RS, Cole DJ (2011) A review of mathematical models of
human sensory dynamics relevant to the steering task. In: Iwnicki
S, Goodall R, Mei TX (eds) The International Association for
123
112
Vehicle System Dynamics, Manchester Metropolitan University,
Manchester, UK
Bigler RS (2013) Automobile driver sensory system modeling. Ph.d.
thesis, Cambridge University
Boer ER (1996) Tangent point oriented curve negotiation. In: Proceedings of conference on intelligent vehicles, IEEE 617:712. doi:10.
1109/IVS.1996.566341
Boer ER (1999) Car following from the drivers perspective. Transp
Res Part F Traffic Psychol Behav 2(4):201206. doi:10.1016/
S1369-8478(00)00007-3
Borah J, Young LR, Curry RE (1988) Optimal estimator model for
human spatial orientation. Ann N Y Acad Sci 545(1):5173.
doi:10.1111/j.1749-6632.1988.tb19555.x
Boring E (1917) A chart of the psychometric function. Am J Psychol
28(4):465470
Bottoms DJ (1983) The interaction of driving speed, steering difficulty and lateral tolerance with particular reference to agriculture.
Ergonomics 26(2):123139. doi:10.1080/00140138308963324
Bremmer F, Lappe M (1999) The use of optical velocities for
distance discrimination and reproduction during visually simulated self motion. Exp Brain Res 127(1):3342. doi:10.1007/
s002210050771
Bronstein AM, Hood JD (1986) The cervico-ocular reflex in normal
subjects and patients with absent vestibular function. Brain Res
373(12):399408. doi:10.1016/0006-8993(86)90355-0
Browning NA, Grossberg S, Mingolla E (2009) A neural model of how
the brain computes heading from optic flow in realistic scenes.
Cogn Psychol 59(4):320356. doi:10.1016/j.cogpsych.2009.07.
002
Burge J, Girshick AR, Banks MS (2010) Visual-haptic adaptation is
determined by relative reliability. J Neurosci Off J Soc Neurosci
30(22):77147721. doi:10.1523/JNEUROSCI.6427-09.2010
Butler JS, Smith ST, Campos JL, Blthoff HH (2010) Bayesian integration of visual and vestibular signals for heading. J Vis 10(11):23.
doi:10.1167/10.11.23
Butler JS, Campos JL, Blthoff HH, Smith ST (2011) The role of
stereo vision in visual-vestibular integration. Seeing Perceiving
24(5):453470. doi:10.1163/187847511X588070
Butler JS, Campos JL, Blthoff HH (2014) Optimal visualvestibular integration under conditions of conflicting intersensory
motion profiles. Exp Brain Res 233(2):587597. doi:10.1007/
s00221-014-4136-1
Campos JL, Blthoff HH (2012) Multimodal integration during selfmotion in virtual reality. In: Murray MM, Wallace MT (eds) The
neural bases of multisensory processes, chap 30. CRC Press, Boca
Raton
Carpenter RHS (1984) Neurophysiology, 4th edn. Arnold, London
Carriot J, Jamali M, Cullen KE (2015) Rapid adaptation of multisensory
integration in vestibular pathways. Front Syst Neurosci 9(59):15.
doi:10.3389/fnsys.2015.00059
Chattington M, Wilson M, Ashford D, Marple-Horvat DE (2007) Eyesteering coordination in natural driving. Exp Brain Res 180(1):1
14. doi:10.1007/s00221-006-0839-2
Clark JJ, Yuille AL (1990) Data fusion for sensory information processing systems. Kluwer Academic Publishers, Norwell
Cloete SR, Wallis G (2009) Limitations of feedforward control in
multiple-phase steering movements. Exp Brain Res 195(3):481
487. doi:10.1007/s00221-009-1813-6
Cole DJ, Pick AJ, Odhams A (2006) Predictive and linear
quadratic methods for potential application to modelling driver
steering control. Veh Syst Dyn 44(3):259284. doi:10.1080/
00423110500260159
Cole DJ (2012) A path-following driver-vehicle model with neuromuscular dynamics, including measured and simulated responses to
a step in steering angle overlay. Veh Syst Dyn 50(4):573596.
doi:10.1080/00423114.2011.606370
123
Biol Cybern (2016) 110:91116

Collins DF, Refshauge KM, Todd G, Gandevia SC (2005) Cutaneous
receptors contribute to kinesthesia at the index finger, elbow, and
knee. J Neurophysiol 94(3):16991706. doi:10.1152/jn.00191.
2005
Correia Grcio BJ, van Paasen MM, Mulder M (2010) Tuning of the
lateral specific force gain based on human motion perception in the
Desdemona simulator . In: AIAA modeling and simulation technologies conference, August, p e8094. doi:10.2514/6.2010-8094
Correia Grcio BJ, Valente Pais AR, van Paassen MM, Mulder M,
Kelly LC, Houck JA (2013) Optimal and coherence zone comparison within and between flight simulators. J Aircr 50(2):493507.
doi:10.2514/1.C031870
Correia Grcio BJ, Bos JE, van Paassen MM, Mulder M (2014) Perceptual scaling of visual and inertial cues: effects of field of view,
image size, depth cues, and degree of freedom. Exp Brain Res
232(2):637646. doi:10.1007/s00221-013-3772-1
de Bruyn B, Orban GA (1988) Human velocity and direction discrimination measured with random dot patterns. Vis Res 28(12):1323
1335. doi:10.1016/0042-6989(88)90064-8
de Winkel K, Soyka F, Barnett-Cowan M, Blthoff HH, Groen EL,
Werkhoven P (2013) Integration of visual and inertial cues in the
perception of angular self-motion. Exp Brain Res 231(2):209218.
doi:10.1007/s00221-013-3683-1
de Winkel KN, Katliar M, Blthoff HH (2015) Forced fusion in multisensory heading estimation. Plos One 10(5):e0127104. doi:10.
1371/journal.pone.0127104
de Winkel K, Correia Grcio BJ, Groen EL, Werkhoven P (2010) Visual
inertial coherence zone in the perception of heading. In: AIAA
modeling and simulation technologies conference, American Institute of Aeronautics and Astronautics, Reston, Virigina, August, p
e7916. doi:10.2514/6.2010-7916
Defazio K, Wittman D, Drury CG (1992) Effective vehicle width in
self-paced tracking. Appl Ergon 23(6):382386
Deneve S, Latham PE, Pouget A (1999) Reading population codes: a
neural implementation of ideal observers. Nat Neurosci 2(8):740
745. doi:10.1038/11205
Dichgans J, Brandt T (1978) Visualvestibular interaction: effects
on self-motion perception and postural control. In: Held R,
Leibowitz HW, Teuber H (eds) Perception. Springer Berlin
Heidelberg, Berlin, Heidelberg, pp 755804. doi:10.1007/
978-3-642-46354-9_25
Donges E (1978) A two-level model of driver steering behavior. Hum
Factors J Hum Factors Ergon Soc 20(6):691707
dos Santos Buinhas L, Correia Grcio BJ, Valente Pais AR, van Paassen
MM, Mulder M (2013) Modeling coherence zones in flight simulation during yaw motion. In: AIAA modeling and simulation
technologies (MST) conference, American Institute of Aeronautics and Astronautics, Reston, Virginia. doi:10.2514/6.2013-5223
Drop FM, Pool DM, Damveld HJ, van Paassen MM, Mulder M (2013)
Identification of the feedforward component in manual control
with predictable target signals. IEEE Trans Cybern 43(6):1936
1949. doi:10.1109/TSMCB.2012.2235829
Drugowitsch J, DeAngelis GC, Klier EM, Angelaki DE, Pouget A
(2014) Optimal multisensory decision-making in a reaction-time
task. ELife 3:e03005. doi:10.7554/eLife.03005
Einstein A (1907) ber die Mglichkeit einer neuen Prfung des Relativittsprinzips. Ann der Phys 328(6):197198. doi:10.1002/andp.
19073280613
Ellerbroek J, Stroosma O, Mulder M, van Paassen MM (2008) Role
identification of yaw and sway motion in helicopter yaw control
tasks. J Aircr 45(4):12751289. doi:10.2514/1.34513
Ernst MO, Banks MS (2002) Humans integrate visual and haptic information in a statistically optimal fashion. Nature 415(6870):429
433. doi:10.1038/415429a
Ernst MO, Blthoff HH (2004) Merging the senses into a robust percept.
Trends Cogn Sci 8(4):162169. doi:10.1016/j.tics.2004.02.002
Biol Cybern (2016) 110:91116

Ernst MO, Di Luca M (2011) Multisensory perception: from integration
to remapping. Sensory cue integration. Oxford University Press,
Oxford, pp 224250
Fernandez C, Goldberg JM (1971) Physiology of peripheral neurons
innervating semicircular canals of the squirrel monkey. Parts I to
III. J Neurophysiol 34(4):661675
Fernandez C, Goldberg JM (1976) Physiology of peripheral neurons
innervating otolith organs of the squirrel monkey. Parts I to III. J
Neurophysiol 39(5):970984
Fetsch CR, Turner AH, DeAngelis GC, Angelaki DE (2009) Dynamic
reweighting of visual and vestibular cues during self-motion perception. J Neurosci Off J Soc Neurosci 29(49):1560115612.
doi:10.1523/JNEUROSCI.2574-09.2009
Fetsch CR, Deangelis GC, Angelaki DE (2010) Visual-vestibular cue
integration for heading perception: applications of optimal cue
integration theory. Eur J Neurosci 31(10):17211729. doi:10.
1111/j.1460-9568.2010.07207.x
Fitzpatrick RC, Day BL (2004) Probing the human vestibular system
with galvanic stimulation. J Appl Physiol (Bethesda, Md : 1985)
96(6):23012316. doi:10.1152/japplphysiol.00008.2004
Flach JM (1990) Control with an eye for perception: precursors to
an active psychophysics. Ecol Psychol 2(2):83111. doi:10.1207/
s15326969eco0202_1
Gawthrop P, Loram I, Lakie M, Gollee H (2011) Intermittent control: a
computational theory of human control. Biol Cybern 104(12):31
51. doi:10.1007/s00422-010-0416-4
Ghahramani Z, Wolpert D, Jordan MI (1997) Computational models of
sensorimotor integration. Adv Psychol 119:117147
Gianna C, Heimbrand S, Gresty M (1996) Thresholds for detection of
motion direction during passive lateral whole-body acceleration
in normal subjects and patients with bilateral loss of labyrinthine
function. Brain Res Bull 40(56):443449
Gibson JJ (1950) Percept Vis World. Houghton Mifflin, Boston
Girshick AR, Banks MS (2009) Probabilistic combination of slant information: weighted averaging and robustness as optimal percepts. J
Vis 9(9):e8. doi:10.1167/9.9.8
Gordon DA (1965) Static and dynamic visual fields in human space
perception. Josa 372(10):12961303
Gordon DA (1966) Experimental isolation of the drivers visual input.
Hum Factors J Hum Factors Ergon Soc 8(2):129138
Gordon T, Lidberg M (2015) Automated driving and autonomous functions on road vehicles. Veh Syst Dyn 53(7):958994. doi:10.1080/
00423114.2015.1037774
Grabherr L, Nicoucar K, Mast FW, Merfeld DM (2008) Vestibular
thresholds for yaw rotation about an earth-vertical axis as a function of frequency. Exp Brain Res 186(4):677681. doi:10.1007/
s00221-008-1350-8
Grant PR, Lee PT (2007) Motion-visual phase-error detection in a flight
simulator. J Aircr 44(3):927935. doi:10.2514/1.25807
Grewal MS, Andrews AP (2001) Kalman filtering: theory and practice
using MATLAB, 2nd edn. Wiley, New York
Groen EL, Wentink M, Valente Pais AR, Mulder M, van Paassen MM
(2006) Motion perception thresholds in flight simulation. In: AIAA
modeling and simulation technologies conference and exhibit,
American Institute of Aeronautics and Astronautics, Reston, Virigina, p e6254. doi:10.2514/6.2006-6254
Groen EL, Bles W (2004) How to use body tilt for the simulation of
linear self motion. J Vestib Res Equilib Orientat 14(5):375385
Grossberg S, Mingolla E, Viswanathan L (2001) Neural dynamics of
motion integration and segmentation within and across apertures.
Vis Res 41(19):25212553
Grunwald AJ, Merhav SJ (1976) Vehicular control by visual field cuesanalytical model and experimental validation. IEEE Trans Syst
Man Cybern 6(12):835845. doi:10.1109/TSMC.1976.4309480
113
Gu Y, Angelaki DE, Deangelis GC (2008) Neural correlates of
multisensory cue integration in macaque MSTd. Nat Neurosci
11(10):12011210. doi:10.1038/nn.2191
Hmlinen M, Hari R, Ilmoniemi RJ, Knuutila J, Lounasmaa
OV (1993) Magnetoencephalography-theory, instrumentation, and
applications to noninvasive studies of the working human brain.
Rev Mod Phys 65(2):413497. doi:10.1103/RevModPhys.65.413
Harris LR, Jenkin M, Zikovitz DC (2000) Visual and non-visual cues in
the perception of linear self-motion. Exp Brain Res 135(1):1221.
doi:10.1007/s002210000504
Heerspink H, Berkouwer W, Stroosma O, van Paassen MM, Mulder M,
Mulder B (2005) Evaluation of vestibular thresholds for motion
detection in the SIMONA research simulator. In: AIAA modeling and simulation technologies conference and exhibit, American
Institute of Aeronautics and Astronautics, Reston, Virigina, p
e6502. doi:10.2514/6.2005-6502
Herrin GD, Neuhardt JB (1974) An empirical model for automobile
driver horizontal curve negotiation. Hum Factors 16(2):129133.
doi:10.1177/001872087401600203
Hosman RJ (1996) Pilots perception and control of aircraft motions.
Ph.d. Thesis, Delft University of Technology
Hosman RJ, Stassen H (1999) Pilots perception in the control of aircraft motions. Control Eng Pract 7(11):14211428. doi:10.1016/
S0967-0661(99)00111-2
Hosman RJ, Van Der Vaart JC (1978) Vestibular models and thresholds
of motion perception: results of tests in a flight simulator. Tech.
rep. Delft University of Technology, Delft
Houck JA, Simon W (1967) Responses of Golgi tendon organs to forces
applied to muscle tendon. J Neurophysiol 30(6):14661481
Ishida M, Fushiki H, Nishida H, Watanabe Y (2008) Self-motion perception during conflicting visualvestibular acceleration. J Vestib
Res Equilib orientat 18:267272
Johansson G (1975) Visual motion perception. Sci Am 232(6):7688.
doi:10.1038/scientificamerican0675-76
Johns TA, Cole DJ (2015) Measurement and mathematical model of a
drivers intermittent compensatory steering control. Vehicle system dynamics (In press) NVSD110,074. doi:10.1080/02699931.
2011.628301
Jonik P, Valente Pais AR, van Paassen MM, Mulder M (2011) Phase
coherence zones in flight simulation. In: AIAA modeling and simulation technologies conference, American Institute of Aeronautics
and Astronautics, Reston, Virigina, August, p e6555. doi:10.2514/
6.2011-6555
Kaliuzhna M, Prsa M, Gale S, Lee SJ, Blanke O (2015) Learning to
integrate contradictory multisensory self-motion cue pairings. J
Vis 15(1):e10. doi:10.1167/15.1.10
Kalman RE (1960) A new approach to linear filtering and prediction
problems. J Basic Eng 82(1):3545. doi:10.1115/1.3662552
Kandel E, Schwartz J, Jessell T (2000) Principles of neural science, 4th
edn. McGraw-Hill, New York
Kandil FI, Rotter A, Lappe M (2009) Driving is smoother and more
stable when using the tangent point. J Vis 9(2009):11.111.11.
doi:10.1167/9.1.11
Kandil FI, Rotter A, Lappe M (2010) Car drivers attend to different gaze targets when negotiating closed vs. open bends. J Vis
10(2010):24.124.11. doi:10.1167/10.4.24
Kawakami O, Kaneoke Y, Maruyama K, Kakigi R, Okada T, Sadato N,
Yonekura Y (2002) Visual detection of motion speed in humans:
spatiotemporal analysis by fMRI and MEG. Hum Brain Mapp
118:104118. doi:10.1002/hbm.10033
Keen SD, Cole DJ (2011) Application of time-variant predictive control
to modelling driver steering skill. Veh Syst Dyn 49(4):527559.
doi:10.1080/00423110903551626
Keen SD, Cole DJ (2012) Bias-free identification of a linear modelpredictive steering controller from measured driver steering behav-
123
114
ior. IEEE Trans Syst Man Cybern Part B Cybern 42(2):434443.
doi:10.1109/TSMCB.2011.2167509
Kingma H (2005) Thresholds for perception of direction of linear acceleration as a possible evaluation of the otolith function. BMC Ear
Nose Throat Disord. doi:10.1186/1472-6815-5-5
Knill DC (2007) Robust cue integration: a Bayesian model and evidence
from cue-conflict studies with stereoscopic and figure cues to slant.
J Vis 7(7):e5. doi:10.1167/7.7.5
Koenderink JJ (1986) Optic flow. Vis Res 26(1):161179. doi:10.1016/
0042-6989(86)90078-7
Kondoh T, Yamamura T, Kitazaki S, Kuge N, Boer ER (2008) Identification of visual cues and quantification of drivers perception
of proximity risk to the lead vehicle in car-following situations. J
Mech Syst Transp Logist 1(2):170180. doi:10.1299/jmtl.1.170
Krding KP, Beierholm U, Ma WJ, Quartz S, Tenenbaum JB, Shams
L (2007) Causal inference in multisensory perception. PloS One
2(9):e943. doi:10.1371/journal.pone.0000943
Lam K, Kaneoke Y, Gunji A, Yamasaki H, Matsumoto E, Naito T,
Kakigi R (2000) Magnetic response of human extrastriate cortex
in the detection of coherent and incoherent motion. Neuroscience
97(1):110. doi:10.1016/S0306-4522(00)00037-3
Land MF (1998) The visual control of steering. In: Harris LR, Jenkin M
(eds) Vision and action, Cambridge University Press, Cambridge,
UK, pp 163180
Land MF, Horwood J (1995) Which parts of the road guide steering?
Nature 377(6547):339430. doi:10.1038/377339a0
Land MF, Lee DN (1994) Where we look when we steer. Nature
369(6483):742744. doi:10.1038/369742a0
Land MF, Tatler BW (2001) Steering with the head. The visual strategy
of a racing driver. Current Biol CB 11(15):12151220
Landy MS, Maloney LT, Johnston EB, Young M (1995) Measurement
and modeling of depth cue combination: in defense of weak fusion.
Vis Res 35(3):389412
Lappe M, Bremmer F, Van Den Berg aV (1999) Perception of selfmotion from visual flow. Trends Cogn Sci 3(9):329336. doi:10.
1016/S1364-6613(99)01364-9
Lappi O, Lehtonen E, Pekkanen J, Itkonen T (2013) Beyond the tangent
point: gaze targets in naturalistic driving. J Vis 13(13):11. doi:10.
1167/13.13.11
Legge GE, Campbell F (1981) Displacement detection in human vision.
Vis Res 21(2):205213. doi:10.1016/0042-6989(81)90114-0
Lehtonen E, Lappi O, Kotkanen H, Summala H (2013) Look-ahead
fixations in curve driving. Ergonomics 56(1):3444. doi:10.1080/
00140139.2012.739205
Levitt H (1970) Transformed updown method in psychoacoustics. J
Acoust Soc Am 49(2b):467477. doi:10.1121/1.1912375
Ljung L (1999) System identification: theory for the user, 2nd edn.
Prentice Hall, Upper Saddle River
Lot R, Dal Bianco N (2015) Lap time optimisation of a racing go-kart.
Veh Syst Dyn 3114(January):121. doi:10.1080/00423114.2015.
1125514
Macadam CC (1981) Application of an optimal preview control for
simulation of closed-loop automobile driving. IEEE Trans Syst
Man CybernSMC 11(6):393399
Macadam CC (2003) Understanding and modeling the human
driver. Veh Syst Dyn 40(13):101134. doi:10.1076/vesd.40.1.
101.15875
MacNeilage PR, Banks MS, Berger DR, Blthoff HH (2007) A
Bayesian model of the disambiguation of gravitoinertial force
by visual cues. Exp Brain Res 179(2):263290. doi:10.1007/
s00221-006-0792-0
Mallery RM, Olomu OU, Uchanski RM, Militchin VA, Hullar TE
(2010) Human discrimination of rotational velocities. Exp Brain
Res 204(1):1120. doi:10.1007/s00221-010-2288-1
123
Biol Cybern (2016) 110:91116

Maltenfort MG, Burke RE (2003) Spindle model responsive to mixed
fusimotor inputs and testable predictions of beta feedback effects.
J Neurophysiol 89(5):27972809. doi:10.1152/jn.00942.2002
Mars F (2008) Driving around bends with manipulated eye-steering
coordination. J Vis 8(11):10.110.11. doi:10.1167/8.11.10
Matthews PB (1984) Evidence from the use of vibration that the human
long-latency stretch reflex depends upon spindle secondary afferents. J Physiol 348:383415
Mileusnic MP, Brown IE, Lan N, Loeb GE (2006) Mathematical models of proprioceptors. I. Control and transduction in the muscle
spindle. J Neurophysiol 96(4):17721788. doi:10.1152/jn.00868.
2005
Mileusnic MP, Loeb GE (2006) Mathematical models of proprioceptors.
II. Structure and function of the Golgi tendon organ. J Neurophysiol 96(4):17891802. doi:10.1152/jn.00869.2005
Mingolla E (2003) Neural models of motion integration and segmentation. Neural Netw Off J Int Neural Netw Soc 16(56):939945.
doi:10.1016/S0893-6080(03)00099-6
Monen J, Brenner E (1994) Detecting changes in ones own velocity
from the optic flow. Perception 23:681690
Nakayama K (1985) Biological image motion processing: a review. Vis
Res 25(5):625660. doi:10.1016/0042-6989(85)90171-3
Naseri AR, Grant PR (2012) Human discrimination of translational accelerations. Exp Brain Res 218(3):455464. doi:10.1007/
s00221-012-3035-6
Nash CJ, Cole DJ (2015) Development of a novel model of drivervehicle steering control incorporating sensory dynamics. In: 24th
International symposium on dynamics of vehicles on roads and
tracks, Taylor & Francis, Graz, Austria
Nesti A, Barnett-Cowan M, MacNeilage PR, Blthoff HH (2014a)
Human sensitivity to vertical self-motion. Exp Brain Res
232(1):303314. doi:10.1007/s00221-013-3741-8
Nesti A, Beykirch KA, Pretto P, Blthoff HH (2014b) Self-motion
sensitivity to visual yaw rotations in humans. Exp Brain Res
233(3):861869. doi:10.1007/s00221-014-4161-0
Nesti A, Beykirch Ka, Pretto P, Blthoff HH (2015) Human discrimination of head-centred visual-inertial yaw rotations. Exp Brain Res.
doi:10.1007/s00221-015-4426-2
Newberry AC, Griffin MJ, Dowson M (2007) Driver perception
of steering feel. Proc Inst Mech Eng Part D J Automob Eng
221(4):405415. doi:10.1243/09544070JAUTO415
Nieuwenhuizen FM, Mulder M, van Paassen MM, Blthoff HH (2013)
Influences of simulator motion system characteristics on pilot control behavior. J Guid Control Dyn 36(3):667676. doi:10.2514/1.
59257
Odhams AMC, Cole DJ (2004) Models of driver speed choice in
curves. In: 7th international symposium on advanced vehicle control (AVEC 04), pp 16
Odhams aMC, Cole DJ (2009) Application of linear preview control to modelling human steering control. Proc Inst Mech
Eng Part D J Automob Eng 223(7):835853. doi:10.1243/
09544070JAUTO1039
Odhams AM, Cole DJ (2014) Identification of the steering control
behaviour of five test subjects following a randomly curving path
in a driving simulator. Int J Veh Auton Syst 12(1):44. doi:10.1504/
IJVAS.2014.057863
Ohmi M (1996) Egocentric perception through interaction among many
sensory systems. Cogn Brain Res 5(12):8796. doi:10.1016/
S0926-6410(96)00044-4
Onur C (2014) Developing a computational model of the pilots best
possible expectation of aircraft state given vestibular and visual
cues. Masters thesis, Georgia Institute of Technology
Oru I, Maloney LT, Landy MS (2003) Weighted linear cue combination
with possibly correlated error. Vis Res 43(23):24512468. doi:10.
1016/S0042-6989(03)00435-8
Biol Cybern (2016) 110:91116

Peng H (2002) Evaluation of driver assistance systemsa human centered approach. In: Proceedings of 6th symposium on advanced
vehicle control
Pick AJ, Cole DJ (2007) Dynamic properties of a drivers arms holding a steering wheel. Proc Inst Mech Eng Part D J Automob Eng
221(12):14751486. doi:10.1243/09544070JAUTO460
Pick AJ, Cole DJ (2008) A mathematical model of driver steering control including neuromuscular dynamics. J Dyn Syst Meas Control
130(3):031004. doi:10.1115/1.2837452
Plchl M, Edelmann J (2007) Driver models in automobile dynamics application. Veh Syst Dyn 45(78):699741. doi:10.1080/
00423110701432482
Pool DM, Mulder M, van Paassen MM, Van Der Vaart JC (2008) Effects
of peripheral visual and physical motion cues in roll-axis tracking
tasks. J Guid Control Dyn 31(6):16081622. doi:10.2514/1.36334
Pool DM, Valente Pais AR, De Vroome AM, van Paassen MM, Mulder
M (2012) Identification of nonlinear motion perception dynamics
using time-domain pilot modeling. J Guid Control Dyn 35(3):749
763. doi:10.2514/1.56236
Poppele RE, Bowman RJ (1970) Quantitative description of linear behavior of mammalian muscle spindles. J Neurophysiol
33(1):5972
Pouget A, Dayan P, Zemel R (2000) Information processing with
population codes. Nat Rev Neurosci 1(2):125132. doi:10.1038/
35039062
Pretto P, Nesti A, Nooij S, Losert M, Blthoff HH (2014) Variable
roll-rate perception in driving simulation. In: Driving simulation
conference,vol 3(1), pp 17
Prochazka A (1999) Quantifying proprioception. Prog Brain Res
123:133142
Prokop G (2001) Modeling human vehicle driving by model predictive
online. Optimization. doi:10.1076/vesd.35.1.19.5614
Proske U, Gandevia SC (2009) The kinaesthetic senses. J Physiol
587(17):41394146. doi:10.1113/jphysiol.2009.175372
Prsa M, Gale S, Blanke O (2012) Self-motion leads to mandatory cue
fusion across sensory modalities. J Neurophysiol 108(8):2282
2291. doi:10.1152/jn.00439.2012
Rader AA, Oman CM, Merfeld DM (2011) Perceived tilt and translation
during variable-radius swing motion with congruent or conflicting
visual and vestibular cues. Exp Brain Res 210(2):173184. doi:10.
1007/s00221-011-2612-4
Raudies F, Neumann H (2012) A review and evaluation of methods
estimating ego-motion. Comput Vis Image Underst 116(5):606
633. doi:10.1016/j.cviu.2011.04.004
Reymond G, Kemeny A, Droulez J, Berthoz A (2001) Role of lateral
acceleration in curve driving: driver model and experiments on a
real vehicle and a driving simulator. Hum Factors J Hum Factors
Ergon Soc 43(3):483495. doi:10.1518/001872001775898188
Riemersma JB (1981) Visual control during straight road driving. Acta
Psychol (Amst) 48(13):215225
Ritchie ML, Mccoy WK, Welde WL (1968) A study of the relation
between forward velocity and lateral acceleration in curves during
normal driving. Hum Factors J Hum Factors Ergon Soc 10(3):255
258. doi:10.1177/001872086801000307
Robertshaw KD, Wilkie RM (2008) Does gaze influence steering around
a bend? J Vis 8(4):18.118.13. doi:10.1167/8.4.18
Rock I, Victor J (1964) Vision and touch: an experimentally created
conflict between the two senses. Science 143(3606):594596
Rodchenko V, Boris S, White A (2000) In-flight estimation of pilots
acceleration sensitivity thresholds. In: Modeling and simulation
technologies conference, American Institute of Aeronautics and
Astronautics, Reston, Virigina, p e4292. doi:10.2514/6.2000-4292
Samji A, Reid LD (1992) The detection of low-amplitude yawing
motion transients in a flight simulator. IEEE Trans Syst Man
Cybern 22(2):300306. doi:10.1109/21.148432
115
Scarfe P, Hibbard PB (2011) Statistically optimal integration of biased
sensory estimates. J Vis 11(7):e12. doi:10.1167/11.7.12
Seilheimer RL, Rosenberg A, Angelaki DE (2014) Models and
processes of multisensory cue combination. Curr Opin Neurobiol
25:3846. doi:10.1016/j.conb.2013.11.008
Sharp RS, Valtetsiotis V (2001) Optimal preview car steering control.
Veh Syst Dyn Suppl 35:101117
Shinar D, Mcdowell ED, Rockwell TH (1977) Eye movements in curve
negotiation. Hum Factors J Hum Factors Ergon Soc 19(1):6371.
doi:10.1177/001872087701900107
Soyka F, Robuffo Giordano P, Beykirch KA, Blthoff HH (2011)
Predicting direction detection thresholds for arbitrary translational acceleration profiles in the horizontal plane. Exp Brain Res
209(1):95107. doi:10.1007/s00221-010-2523-9
Soyka F, Giordano PR, Barnett-Cowan M, Blthoff HH (2012) Modeling direction discrimination thresholds for yaw rotations around
an earth-vertical axis for arbitrary motion profiles. Exp Brain Res
220(1):8999. doi:10.1007/s00221-012-3120-x
Soyka F, Blthoff HH, Barnett-Cowan M (2013) Temporal processing of
self-motion: modeling reaction times for rotations and translations.
Exp Brain Res 228(1):5162. doi:10.1007/s00221-013-3536-y
Soyka F, Blthoff HH, Barnett-Cowan M (2015) Integration of semicircular canal and otolith cues for direction discrimination during eccentric rotations. Plos One 10(8):e0136,925. doi:10.1371/
journal.pone.0136925
Soyka F, Teufel H, Beykirch KA, Robuffo Giordano P, Butler JS,
Nieuwenhuizen FM, Blthoff HH (2009) Does jerk have to be considered in linear motion simulation? In: Proceedings of the AIAA
modeling and simulation technologies conference, pp 13811388.
doi:10.2514/6.2009-6245
Steen J, Damveld HJ, Happee R, van Paassen MM, Mulder M (2011) A
review of visual driver models for system identification purposes.
In: 2011 IEEE international conference on systems, man, and
cybernetics, pp 20932100. doi:10.1109/ICSMC.2011.6083981
Steinhausen W (1933) ber die Beobachtung der Cupula in den Bogengangsampullen des Labyrinths des lebenden Hechts. Pflgers
Archiv fr die Gesamte Physiologie des Menschen und der Tiere
232(1):500512. doi:10.1007/BF01754806
Tabak S, Collewijn H, Boumans LM, Van Der Steen J (1997) Gain
and delay of human vestibulo-ocular reflexes to oscillation and
steps of the head by a reactive torque helmet. Acta Oto-laryngol
117(6):785795. doi:10.3109/00016489709114203
Telban RJ, Cardullo F (2005) Motion cueing algorithm development:
human-centered linear and nonlinear approaches. NASA technical
report. Nasa Langley Research Center, Hampton, VA, USA
Thommyppillai M, Evangelou S, Sharp RS (2009) Car driving at the
limit by adaptive linear optimal preview control. Veh Syst Dyn
47(12):15351550. doi:10.1080/00423110802673109
Timings JP, Cole DJ (2013) Minimum maneuver time calculation using
convex optimization. J Dyn Syst Meas Control 135(3):e031015.
doi:10.1115/1.4023400
Timings JP, Cole DJ (2014) Robust lap-time simulation. Proc Inst Mech
Eng Part D J Automob Eng 228(10):12001216. doi:10.1177/
0954407013516102
Todorov E (2005) Stochastic optimal control and estimation methods adapted to the noise characteristics of the sensorimotor system. Neural Comput 17(5):10841108. doi:10.1162/
0899766053491887
Trojaborg W, Sindrup EH (1969) Motor and sensory conduction in
different segments of the radial nerve in normal subjects. J Neurol
Neurosurg Psychiatry 32(4):354359
Ullman S (1979) Interpretation visual motion. Massachusetts Inst of
Technology Pr, Oxford
Ungoren A, Peng H (2005) An adaptive lateral preview
driver model. Veh Syst Dyn 43(4):245259. doi:10.1080/
00423110412331290419
123
116
Vaitl D, Mittelstaedt H, Saborowski R, Stark R, Baisch F (2002) Shifts
in blood volume alter the perception of posture: Further evidence
for somatic graviception. Int J Psychophysiol 44(1):111. doi:10.
1016/S0167-8760(01)00184-2
Valente Pais AR, Mulder M, van Paassen MM, Wentink M, Groen
EL (2006) Modeling human perceptual thresholds in self-motion
perception. In: AIAA modeling and simulation technologies
conference and exhibit, American Institute of Aeronautics and
Valente Pais AR, van Paassen MM, Mulder M, Wentink M (2010b)
Perception of combined visual and inertial low-frequency yaw
motion. In: AIAA modeling and simulation technologies conference, American Institute of Aeronautics and Astronautics, Reston,
Virigina, August, pp 110. doi:10.2514/6.2010-8093
Valente Pais AR, van Paassen MM, Mulder M, Wentink M (2011)
Effect of performing a boundary-avoidance tracking task on the
perception of coherence between visual and inertial cues. In: AIAA
modeling and simulation technologies conference, American Institute of Aeronautics and Astronautics, Reston, Virigina, August, p
e6324. doi:10.2514/6.2011-6324
Valente Pais AR, van Paassen MM, Mulder M, Wentink M (2010a) Perception coherence zones in flight simulation. J Aircr 47(6):2039
2048. doi:10.2514/1.C000281
Valente Pais AR, Pool DM, de Vroome AM, van Paassen MM, Mulder
M (2012) Pitch motion perception thresholds during passive and
active tasks. J Guid Control Dyn 35(3):904918. doi:10.2514/1.
54987
Valko Y, Lewis RF, Priesol aJ, Merfeld DM (2012) Vestibular labyrinth
contributions to human whole-body motion discrimination. J Neurosci 32(39):13,53713,542. doi:10.1523/JNEUROSCI.2157-12.
2012
van der Steen HFAM (1998) Self-motion perception. Ph.d. thesis, Delft
University of Technology, The Netherlands
Vaseghi S (2005) Advanced digital signal processing and noise reduction, vol 9, 3rd edn. Wiley, Chichester. doi:10.1002/0470094966
Wallis G, Chatziastros A, Blthoff HH (2002) An unexpected role for
visual feedback in vehicle steering control. Curr Biol 12(4):295
299. doi:10.1016/S0960-9822(02)00685-1
Weber E (1834) Annotationes anatomicae et physiologicae. CF Koehler,
Leipzig
Wentink M, Correia Grcio BJ, Bles W (2009) Frequency dependence
of allowable differences in visual and vestibular motion cues in a
simulator. In: AIAA modeling and simulation technologies conference, American Institute of Aeronautics and Astronautics, Reston,
Virigina, August, p e6248. doi:10.2514/6.2009-6248
Wierwille WW, Casali JG, Repa BS (1983) Driver steering reaction
time to abrupt-onset crosswinds, as measured in a moving-base
driving simulator. Hum Factors 25(1):103116
Wolpert DM, Diedrichsen J, Flanagan JR (2011) Principles of sensorimotor learning. Nat Rev Neurosci 12:739751. doi:10.1038/
nrn3112
Wolpert DM, Ghahramani Z (2000) Computational principles of movement neuroscience. Nat Neurosci Suppl 3:12121217. doi:10.
1038/81497
Wright WG, DiZio P, Lackner JR (2005) Vertical linear self-motion
perception during visual and inertial motion: more than weighted
summation of sensory inputs. J Vestib Res Equilib Orientat
15(4):185195
123
Biol Cybern (2016) 110:91116

Young LR (2011) Optimal estimator models for spatial orientation and
vestibular nystagmus. Exp Brain Res 210(34):465476. doi:10.
1007/s00221-011-2595-1
Young LR, Meiry JL (1968) A revised dynamic otolith model. Aerosp
Med 39:606608
Young LR, Oman CM (1969) Model for vestibular adaptation to horizontal rotation. Aerosp Med 40(10):10761080
Yuille AL, Blthoff HH (1996) Bayesian decision theory and psychophysics. In: Richards DCK, Whitman (eds) Perception as
Bayesian inference. Cambridge University Press, Cambridge, pp
123162
Zaal PMT, Pool DM, de Bruin J, Mulder M, van Paassen MM (2009a)
Use of pitch and heave motion cues in a pitch control task. J Guid
Control Dyn 32(2):366377. doi:10.2514/1.39953
Zaal PMT, Pool DM, Chu QP, Mulder M, van Paassen MM, Mulder
JA (2009b) Modeling human multimodal perception and control
using genetic maximum likelihood estimation. J Guid Control Dyn
32(4):10891099. doi:10.2514/1.42843
Zaal PMT, Pool DM, Mulder M, van Paassen MM (2009c) Multimodal
pilot control behavior in combined target-following disturbancerejection tasks. J Guid Control Dyn 32(5):14181428. doi:10.
2514/1.44648
Zaal PMT, Pool DM, Mulder M, van Paassen MM, Mulder JA (2010)
Identification of multimodal pilot control behavior in real flight. J
Guid Control Dyn 33(5):15271538. doi:10.2514/1.47908
Zaal PMT, Pool DM, Paassen MV, Mulder M (2012) Comparing multimodal pilot pitch control behavior between simulated and real
flight. J Guid Control Dyn 35(5):14561471. doi:10.2514/1.56268
Zaal PMT, Nieuwenhuizen FM, van Paassen MM, Mulder M (2013)
Modeling human control of self-motion direction with optic flow
and vestibular motion. IEEE Trans Syst Man Cybern 43(2):544
556. doi:10.1109/TSMCB.2012.2212188
Zacharias GL (1977) Motion sensation dependence on visual and
vestibular cues. Ph.d. thesis, Massachusetts Institute of Technology, Cambridge, MA
Zacharias GL, Caglayan AK, Sinacori JB (1985) A model for visual
flow-field cueing and self-motion estimation. IEEE Trans Syst Man
Cybern SMC 15(3):385389. doi:10.1109/TSMC.1985.6313373
Zacharias GL, Young LR (1981) Influence of combined visual and
vestibular cues on human perception and control of horizontal rotation. Exp Brain Res 41(4):159171
Zaichik L, Rodchenko V, Rufov I, Yashin Y, White A (1999) Acceleration perception. In: Modeling and simulation technologies
conference and exhibit, American Institute of Aeronautics and
Zaidel A, Turner AH, Angelaki DE (2011) Multisensory calibration
is independent of cue reliability. J Neurosci Off J Soc Neurosci
31(39):1394913962. doi:10.1523/JNEUROSCI.2732-11.2011
Zupan LH, Merfeld DM, Darlot C (2002) Using sensory weighting to
model the influence of canal, otolith and visual cues on spatial orientation and eye movements. Biol Cybern 86(3):209230. doi:10.
1007/s00422-001-0290-1
Transportmetrica B: Transport Dynamics
ISSN: 2168-0566 (Print) 2168-0582 (Online) Journal homepage: http://www.tandfonline.com/loi/ttrb20
A review of system dynamics models applied in

transportation
S.P. Shepherd
To cite this article: S.P. Shepherd (2014) A review of system dynamics models
applied in transportation, Transportmetrica B: Transport Dynamics, 2:2, 83-105, DOI:
10.1080/21680566.2014.916236
Published online: 23 May 2014.
Article views: 1129
View Crossmark data

http://www.tandfonline.com/action/journalInformation?journalCode=ttrb20
Date: 16 June 2016, At: 23:55
Transportmetrica B: Transport Dynamics, 2014

A review of system dynamics models applied in transportation

S.P. Shepherd
Institute for Transport Studies, University of Leeds, Leeds LS2 9JT, UK

(Received 27 January 2014; accepted 14 April 2014)
It is 20 years since Abbas and Bell [1994. System Dynamics Applicability to Transportation
Modeling. Transportation Research Part A 28 (5): 373390] evaluated the strengths and weaknesses of system dynamics (SD) as an approach for modelling in the transportation area. They
listed 12 advantages of the approach and in particular suggested it was well suited to strategic
issues and that it could provide a useful tool for supporting policy analysis and decision-making
in the transport field. This paper sets out a review of over 50 peer-reviewed journal papers since
1994 categorising them by area of application and providing a summary of particular insights
raised. The fields of application include the take-up of alternate fuel vehicles, supply chain
management affecting transport, highway maintenance, strategic policy, airport infrastructure
and airline business cycles and a set of emerging application areas. The paper concludes with
recommendations for future application of the SD approach.
Keywords: system dynamics; transport modelling; transportation
1. Introduction
Abbas and Bell (1994) outlined the modelling approach used in system dynamics (SD) and listed
12 advantages of the approach compared to traditional transport modelling. In particular, they
suggested the approach would be well suited to strategic policy analysis and as a support tool for
decision-making. In essence, transportation systems are complex, they often involve a number
of different stakeholders or agents which results in feedbacks with different time lags between
the responses of each type of user. System dynamics models offer a whole system approach to
transport planning and with this different perspective the importance of these feedbacks and lagged
responses can be demonstrated to policy makers. The SD platforms also offer specialised tools,1
which aid in the calibration of models, optimisation of policies and improve ease of use through
flight simulators which all contribute to the understanding of the whole underlying system.
System dynamics was initially developed by Forrester from MIT in the 19501960s (Forrester
1958) and is a powerful methodology developed from system theory, information science, organisational theory, control theory, tactical decision-making, cybernetics and military games. It uses
a standard causal loop approach to develop qualitative models of a system which may be used to
develop dynamic hypotheses before a quantitative stock-flow model is developed. Early applications were in business management but over the past few decades it has been applied to other areas,
including government policy, healthcare, the automobile industry and urban studies (Sterman
2000). The application of causal loop diagrams (which set out the causal links between concepts)
may be used to bring out the mental models (how people think a system works) of different
stakeholders and therefore help remove any barriers to implementation of a given policy. System
Email:
s.p.shepherd@its.leeds.ac.uk
2014 Hong Kong Society for Transportation Studies Limited
84
S.P. Shepherd
dynamics approaches are becoming increasingly used in a hierarchical manner which allows systems and policies to interact across space and time. The holistic approach is well suited to the
transport problems we now face. Using different modelling approaches can also produce significantly different outcomes on the design of policies. System dynamics can bring in other modelling
structures and help explore the real drivers of future demand as well as explain how to change
user perception and behaviour. Models can be built with stakeholders input and then used in the
form of games or flight simulators for policy learning.
The aim of this review is to set out which areas of transportation research have applied SD
since the paper by Abbas and Bell, and in doing so highlight whether the studies have made
use of the qualitative causal loop approach, the quantitative stock-flow modelling approach and
which insights were possible (if any) over and above a more traditional modelling approach. The
approach taken was to review only peer-reviewed journal articles between 1995 and 2013. Whilst
there are many conference papers applying SD approaches to transport problems, the element of
peer-review was used to check for quality. A limited number of key word search terms including,
transport, transportation, SD, modelling were used and after filtering out via reading the abstracts
over 50 papers are included in the review. The papers are categorised by area of application in
what follows while the first section sets out some basics of the SD approach for those who are
unfamiliar with it. The paper then concludes with a discussion about future application areas.
2. Some basics of system dynamics

The SD approach links qualitative and quantitative models. The qualitative models are best built
with the input of the relevant stakeholders and are generally communicated with causal loop
diagrams (CLDs). The development of a CLD is part of the model building process and connects
entities by causal relationships and as the diagram develops then feedback loops become evident.
These loops are either positive (self-reinforcing) or negative (self-correcting or balancing) feedback loops. Reinforcing loops amplify what is happening in the system, i.e. where an increase
in one parameter leads to an increase in another, and without a balancing loop then this leads
to exponential growth. Balancing loops are relationships that oppose change, so in such a loop
an increase in one entity leads to a decrease in another. When systems contain balancing and
reinforcing loops then a dynamic equilibrium may be reached.
A basic example of a simple reinforcing and balancing loop in a system is shown in 1, adapted
from Sterman (2000). The causal loops represent the interaction between eggs, chickens and road
crossings. The eggs and chickens loop is reinforcing (denoted by R) as more eggs lead to more
chickens, which in turn lead to more eggs. If this loop was operating on its own, both chickens and
eggs would increase exponentially. On the other hand, the chickens and road crossings loop is
balancing (denoted by B); while an increase in chickens leads to an increase in road crossings, the
increase in road crossings leads to fewer chickens due to them being run over (denoted by ). If
this loop were operating alone, the chickens (and road crossings) would gradually decline to zero.
As both loops interact, the path of eggs, chickens and road-crossings over time are dependent
on the relative rates but will eventually reach a dynamic equilibrium. Important elements of the
diagram are the linking arrows with polarity + or assigned to show causal links. A positive sign
is used to show that an increase in the first entity causes an increase in the second entity (all else
held constant). More chickens mean more eggs and more road crossings. A negative sign is used
to show that an increase in the first entity causes a reduction in the second. More road crossings
means less chickens. With more complex loops, a positive feedback loop is one where there are
an even number of negative links (or none); and a negative feedback loop is one where there are
an odd number of negative links.
Figure 1.
85
Eggs, chickens, and road crossings.
Figure 2. The bath tub metaphor for stock and flow with integral equation.
While qualitative models are useful in describing the structure of a system and a dynamic
hypothesis, most decision makers then wish to see some quantitative results. Here the approach is
based on linking differential equations but is presented to the user in terms of stocks and flows
via a stock-flow diagram which keeps the model transparent and easy to understand.
Stocks are accumulations and are represented by rectangles suggesting a box to hold the content.
Flows may be inflow to a stock or an outflow from a stock and are represented by pipes with
valves controlling the rate of flow into or out of a stock. Again taking the example from Sterman
(2000), the hydraulic or bathtub metaphor is used to explain the concept and mathematics behind
the simplest building block of one stock. Figure 2 shows the bathtub where the stock or level of
water accumulates over time with inflow controlled by the tap and outflow by the plug. The system
dynamic representation below shows the symbols used in VENSIM, one of the software platforms
used for developing SD models (others use very similar symbols). Underlying the symbols is the
mathematical notation which shows how the stock is the integral of inflowoutflow starting with
an initial level of stock. While the bath tub metaphor is easy to follow, it is also easy to see how
the stock could represent other systems such as population with births and deaths being the inflow
and outflow from the stock of people.
86
S.P. Shepherd
Table 1.
Categorisation of papers by application areas.
Application area
Modelling the uptake of alternate fuel vehicles

Supply chain management with transportation
Highway maintenance/construction
Strategic policy at urban, regional and national levels
Airlines and airports
Emerging areas
Total
Number of papers
12
6
5
13
10
8
54
As stocks and flows are linked to other stocks and flows, a system structure develops and the
system is described by a series of linked non-linear differential equations. Like all modelling,
developing a good model is more like an art and part of the learning process is to develop
the model along with stakeholders. Where this is not possible, then simple concept models are
often useful to explain endogenous system behaviour and some system dynamicists prefer the
use of small models over the more complex and detailed ones see for example the discussion
in Ghaffarzadegan, Lyneis, and Richardson (2011) who discuss the benefits of using small SD
models when addressing public policy issues.
The following sections review over 50 journal papers which have applied SD to a transportation
problem. The papers have been categorised as in Table 1 which shows the category and number
of papers included in the review. Each section does not review all papers in detail but does draw
out examples of good practice which demonstrate the application of CLD or stock flow models
which provide something different to the more traditional transport modelling approach either in
terms of insight or coverage of the problem. As this is a review it is not feasible to go into the
detailed mathematics behind the stock-flow models, but readers may find this material within the
papers quoted.
3. Modelling the uptake of alternate fuel vehicles

With the recent interest around the world in the promotion of alternative fuel vehicles such as
battery electric, plug-in hybrids and hydrogen fuel cell vehicles, it is not surprising that modelling
their uptake has been a hot topic in the application of system dynamic models. System dynamics
is a good fit to this type of problem as previous studies have investigated technology or product
diffusion in other sectors where the Bass diffusion model has been applied and adapted. Typically,
the quantitative papers include such a diffusion process, a fleet ageing chain and a choice model
for the purchase decision with varying levels of detail or market segmentation. Before going
into the stock flow or quantitative models, Stepp et al. (2009) develop a CLD approach with
stakeholders to investigate potential policy implications of supporting high-efficiency vehicles.
They consider consumer preferences, producer decision making, vehicle market dynamics and
lifecycle environmental impacts including effects of production and recycling. An extract from
their CLD is shown in Figure 3 which they use to show the potential for policy resistance to
purchase subsidies.
Implementing a subsidy for high efficiency vehicles in Figure 3 reduces the market retail price
which increases the share of high efficiency vehicles. This alone would reduce emissions, however
as efficiency is increased then the demand for fuel is reduced which as shown in loop B7 reduces
the cost per mile and therefore increases the vehicle miles travelled. This is the usual rebound
effect. However as the market share increases then there is also a macrolevel effect on the price
of fuel which affects costs per mile for other vehicles and increases miles driven per vehicle.
87
Figure 3. Example feedbacks in response to a high efficiency vehicle subsidy.

Source: Stepp et al. (2009).
This increase in vehicle miles driven may then impact on the scrappage loop R1, increasing
the demand for new vehicles earlier which increases production emissions. Finally, the authors
argue that the increased demand for high efficiency vehicles could via loop B1 (the demandprice effect), increase the price of such vehicles at least in the short term which goes against
the intended impact of the subsidy. Whilst the actual outcome depends on the relative strengths
of these loops or interactions, using such a diagram is useful in communicating the potential
unintended consequences or policy resistance which may occur. The authors also discuss how
policies which increase the cost per mile such as carbon taxes, may achieve synergy with such
subsidies via the CLDs.
88
S.P. Shepherd
Figure 4. Principal feedbacks conditioning willingness to consider new platforms and expected modes of
behaviour.
Source: Struben and Sterman (2008).
Moving on to the stock-flow models, Struben and Sterman (2008) develop a framework for modelling the uptake of alternate fuel vehicles (AFVs) which consists of three main elements: a fleet
turnover or stock model, a discrete choice model of the purchase decision and a social/technology
diffusion process (Figure 4). This process extends the Bass diffusion concept to include the impacts
of word of mouth, marketing and social exposure to the new vehicles. It allows for a development
of the choice set or willingness to consider the option over time including the effect of forgetting
about the new vehicle types which then permits a failing market solution. This is an example of
where SD models can bring something different to the process of policy assessment the possibility of an initial uptake of new technology which then fails and which has been observed in
practice when subsidies for compressed natural gas (CNG) vehicles were removed in Canada and
New Zealand (Flynn 2002). Whilst using an example of adoption of waste recycling, Ulli-Beer
et al. (2010) describe the general structures required to model acceptance dynamics involving
paradigm shift involving endogenous preference and value change, these structures are useful in
modelling acceptance and/or rejection of new technology in many areas.
Other studies incorporate similar diffusion structures but add other policy dimensions such as
regulation, including manufacturer responses, Walther et al. (2010), add other indicators such
as impact on fuel duty revenues (Shepherd, Bonsall, and Harrison 2012) and the co-evolution
of infrastructure for re-fuelling and extending to the cases of natural gas, Jannsen et al. (2006)
and hydrogen fuel cell vehicles (Meyer and Winebrake 2009; Kohler et al. 2010; Park, Kim, and
Lee 2011). Meanwhile others have investigated the role of strategic niche management, Kwon
89
(2012), the implication of policies on distributional issues, Harrison and Shepherd (2014) and
applied similar diffusion models to the case of electric motorcycles in Taiwan (Trappey et al.
2012). In terms of impact of policies, most studies find that the uptake is not affected greatly
by subsidies but more by regulation and infrastructure. However, as both Struben and Sterman
(2008) and Shepherd, Bonsall, and Harrison (2012) point out, the results may be more sensitive
to the assumed strength of the word of mouth or marketing effects rather than to changes in the
technical attributes of the alternate vehicles. This ability to bring in soft issues such as impacts
of social exposure and to quickly demonstrate the sensitivity of results to assume parameters are
another strength of the SD approach.
4. Supply chain management with transportation
Tako and Robinson (2012) provide a full review of discrete event and SD applications in the
logistics and supply chain context between 1996 and 2006. A total of 38 papers from 127 used
the SD approach which has a strong presence in the analysis of supply chains due to the natural
ability to represent the well-known bullwhip effect.2 Of these 38 papers only a few investigate
the impacts of supply chain management on the related transport system as shown in Table 1.
Sterman (2000) developed a generic SD model of the basic stock management structure which
explains the sources of oscillation, amplification and phase lag which are observed in real supply chains. Georgiadis, Vlachos, and Iakovou (2005) use these structures to construct a holistic
model by linking single echelon models with similar structures to represent different elements
of the supply chain. They apply the model to a case study of a fast food chain in Greece to
investigate capacity planning under dynamic growth assumptions and the question of when to
increase fleet size rather than lease extra capacity to meet delivery needs. The model is used to
optimise the number of company owned trucks while minimising total transportation costs. Potter
and Lalwani (2008) investigate a similar problem. Variation in daily demands is made worse by
demand amplification or the bullwhip effect and this can make it difficult for hauliers to judge
investments in their fleet compared with how much to sub-contract. The paper confirms previous
qualitative assessments in that demand amplification impacts adversely on transport costs and
performance/efficiency. However, Potter et al. are also able to identify exceptions when vehicle
capacity is just less than average demand, where an increase in demand amplification can improve
transport performance. This is due to spare capacity being available within the transport system;
however, this strategy comes with penalties incurred as extra vehicles are needed to deal with
average demand.
Disney, Potter, and Gardner (2003), Otto and Kotzab (2003) and Wilson (2007) use the SD
approach to investigate the potential for dealing with the bullwhip effect by using the vendor
managed inventory approach. Figure 5 shows a typical supply chain where times to respond and
estimates of demand for orders vary along the chain. Under the VMI approach the warehouse now
acts as a distribution centre and the tier 1 supplier controls the order process from direct estimates of
demand from the customer base. Information flows in the traditional approach between retailer and
warehouse and warehouse and tier 1 supplier are effectively replaced by this direct connection to
the customer which both improves accuracy of demand forecasts and reduces delays in the system.
The rest of the supply chain is as in the traditional model. Wilson (2007) investigated transportation
disruptions at each point along the supply chain. The most critical point was found to be between
the tier 1 supplier and the warehouse, and this was therefore seen as the most important point in
the supply chain and as the place to start with risk management strategies to mitigate the effect
of transportation disruptions. The simple approach to management of inventory by changing the
information flows was shown to reduce demand amplification and as Wilson (2007) shows VMI
can be used to protect against severe transportation disruptions. VMI protects the retailer against
90
Figure 5.
S.P. Shepherd
Supply chain management structure with VMI connection (adapted from Wilson 2007).
the disruption and they do not suffer a peak in inventory, sharing information essentially smooths
out the disruptive effects. The application of SD to this type of problem including transportation
was a natural extension of the more traditional SD literature on supply chain or stock management.
5. Highway maintenance/construction
There are five papers which deal with the highway maintenance and construction problem. Chasey,
de La Garza, and Drew (2002) and Fallah-Fini et al. (2010) model the impact of deferred maintenance on a highway system. Figure 6 adapted from Fallah-Fini et al. shows the aggregate feedback
loops between deterioration and maintenance. The pavement condition deteriorates due to load
Figure 6. Highway deterioration and maintenance CLD.

Source: Fallah-Fini et al. (2010).
91
factors and climate, etc. The balancing loop B1 (maintenance fix) shows that budget allocated
to maintenance operations increases the highway improvement rate which then decreases the
area of highway under distress which then balances or reduces the desired maintenance budget.
The reinforcing loop R1 (accelerated deterioration), includes the effect of maintenance budget
shortfall which causes a delay in maintenance which increases the highway deterioration rate
counter-acting the balancing loop. Whilst the physics of road deterioration and maintenance are
complex in nature, the corresponding feedback structure is relatively simple. Fallah-Fini et al.
developed a dynamic micro-level simulation model of highway deterioration and renewal processes. The model is calibrated with data from eight road sections in Virginia. The model is then
linked with an optimisation module to improve maintenance operations. The analysis results in
different priority setting schemes that improve on current maintenance practices at both the project
and network level. The approach suggests moving towards less costly preventative maintenance
rather than more expensive (deferred) corrective maintenance should bring in benefits to the system as a whole. Again a move away from the traditional approach to the problem provides insight
and promotes alternative strategies.
Friedman (2006) however, questions the preventative maintenance policy. Friedman models the
effect of road conditions on accident development and uses data to demonstrate that the mental
model behind the policy of maintaining better roads to reduce accidents could in most cases
be increasing accidents due to increased speeds and polishing effects. A more complex mental
and stock-flow model is developed which better fits the evidence of increased accidents with
increased pavement service index after maintenance. This demonstrates the importance of setting
boundaries when viewing problems and the importance of evidence to support the mental models
of basic decisions. Friedman questions the current mental model held within the road authority
and demonstrates the importance of accounting not only for empirical evidence but also for the
reasoning by decision-makers who develop their own mental models of short term consequences
where data are missing.
The other two papers in this area relate to policy in China. Hang and Li (2011) develop a model
to investigate the potential role of truck weight regulation policies for a region in China. While
it includes a fleet turn-over model, it does not make much use of other SD features. However,
the paper claims to bring insight which other models do not e.g. that adjustment to a rigid policy
(which requires a change in trucks) needs time and the disruption causes more problems to society
than a more moderate approach.
Xu, Mashayekhib, and Saeed (1998) develop a SD model to investigate the cost overrun problem
in Chinese highway construction projects. The existing policy was described as maximising
quantity over quality in terms of building many roads at low cost with an average life of only
15 years compared with 40 years for Western Europe. The model shows that as demand for
roads grows then the pressure to complete on time and allocation of budget between capital and
maintenance costs meant that quality was reduced as a result of the system itself. This was due
to the funding mechanism for new roads as revenues come from completed projects via the road
users. This meant that there was a pressure to build faster with lower quality so that more revenue
could be generated to fund more roads. A new equilibrium of project starts evolves but with a lower
quality. The study showed that this problem can be overcome with loan financing and increases in
user charges which releases budget constraints and hence pressures on the construction process
to deliver on time (but with reduced quality).
6. Strategic policy at urban, regional and national levels
This section includes studies of high level or strategic policy at the urban, regional, and national
level. Firstly, there are a number of models which look at implications for the structure of cities
92
S.P. Shepherd
and regions as the economy, population, migration, infrastructure and land use interact with
transportation. A sub-set of these build on the tradition of land use interaction models (LUTI). A
feature of land use transport interaction is that these two systems operate on different time scales.
Transport users may respond relatively quickly to changes in transport policy or costs, typically
within months while the land-use system includes a significant degree of inertia. This is due to the
fact that land use in urban areas depends on physical structures such as buildings and transport
infrastructure. This time dependence between the systems makes it ideal for a SD approach.
Despite this most LUTI models are based on the traditional approach in transport modelling i.e.
the notion of equilibrium and are static in nature. These equilibrium based models are used to
forecast changes in transport demand and land use 20 or 30 years hence with no real connection
from the current conditions to those future projections. The pathway to the future state is therefore
unknown.
Pfaffenbichler, Emberger, and Shepherd (2010) and Pfaffenbichler (2011) introduce the concepts underlying the MARS LUTI model. They cover validation and transferability between cities
and provide some examples of the model applied in practice. The model has now been applied in
more than 20 cities world-wide and has been used as a training tool for planners and practitioners
in Asia and in optimising strategic transport policies for a range of cities/regions. The model is
seen as a departure from the norm in transport planning yet has proved to be a useful tool at the
design stage. In particular, the model was first of all designed using a causal loop analysis. One
example concerns the link between planning decisions about infrastructure (road capacity) and the
symptom of congestion. A simplified view of the CLD reported in Pfaffenbichler (2011) is shown
in Figure 7 along-side one of the common two loop Archetypal Structures from Wolstenholme
(2003) which represents short-term fixes which fail. The left figure shows the common archetype
where a short-term fix to a problem then fails due to some unintended consequence which causes
the problem symptom to grow after some delay. Usually the positive reinforcing effects are outside
of the system boundary considered. The right-hand side shows the case for congestion which in
the past has been solved by planners with increased capacity which then increases the attractiveness of car use which in turn increases congestion again. Pfaffenbichler (2011) provides a more
detailed version of this concept and demonstrates that real-world behaviour follows this pattern
with evidence of traffic flows as infrastructure is developed for cases in Austria over a period
Figure 7.
Fixes that fail archetype and congestion capacity car use problem.
93
of more than 30 years. A similar argument is developed for the interaction between infrastructure for cars and the emergence of urban sprawl. The MARS model has been applied in various
studies to investigate optimal integrated transport and land-use strategies (see references within
Pfaffenbichler, Emberger, and Shepherd 2010) where impacts are assessed more rapidly than with
traditional models. It has also been linked with a fleet model by the same author called SERAPIS
so that choice of car as technologies change may also be studied (Pfaffenbichler, Krutak, and
Renner 2011).
Along similar lines, Haghani, Lee, and Byun (2003a) present a regional model based on an analysis of causal relationships and a feedback loop structure connecting a large number of physical,
socio-economic, and policy variables. The model is made up of seven sub-models: population,
migration, households, employment, residential and commercial development, travel demand,
and traffic supply or congestion effects. The model was programmed in the DYNAMO simulation
language, and based on data from Montgomery county, MD. The model is well documented and
includes calibration analysis using historic data from 1970 to 1980 validated against 19801990.
Haghani, Lee, and Byun (2003b) then report robustness and sensitivity tests covering inputs such
as birth rate and regional economic growth rates. The model was used to assess the impacts of
highway capacity expansion and the related changes in land use which in turn affected demand
and the performance of the transport network.
Shen et al. (2009) develop a high-level model comparing low/high density land use policies
for Hong Kong. In the distant long term they show that compact high density scenarios are more
sustainable with investment in rail-based transport over car infrastructure which is seen to be
more prominent in the low density case. The growth in population in the very long term can be
accommodated only in the high density scenario so the authors conclude: that only by means of
a planning policy scheme to support compact and high-density development could Hong Kong
meet the environmental, social and economical requirements of sustainable land use and achieve
a perfect balance among them.
Wang, Lu, and Peng (2008) again develop a very high-level interaction model between population, vehicle ownership, environment, GDP, travel demand and infrastructure supply, applying it
to a case study of Dalian, China. Car ownership policies are studied and the wider system effects
mean that strict ownership controls results in a larger city with a greater gross domestic product
(GDP) and greater share of public transport. It is noticeable that by 2050 the interplay between
population, car ownership, congestion, pollution and GDP mean that there are a similar number of
vehicles in total but for quite different populations. Cities with vehicle ownership encouraged have
a population of around 3 million versus 9 million inhabitants where ownership is not encouraged,
however it appears that the larger city has a lower GDP per capita.
Feng and Hsieh (2009) propose a hybrid model which integrates SD, cognitive maps, and
a sensitivity model to study the problem of investment in transport systems and in particular the question of resource allocation over time or when to invest. A case study for Taipei is
used to illustrate the approach to satisfying the needs of multiple stakeholders. The results of
the sensitivity analysis revealed that the increase in private vehicle trips caused increases in
emissions, energy consumed and accidents and so failed to meet the competing needs of multiple stakeholders. However, the system is shown to be insensitive to managers decisions on
resource allocation timings i.e. delays to implementation are not the problem. Policies which
control the growth in car use are shown to be the most effective at meeting the needs of
stakeholders.
All the above studies deal with short-term policies which have long-term impacts on urban
form or infrastructure and vice versa. System dynamics is ideal for investigating such systems
which contain feedbacks and delays which are often outside of the mental model of the decision
maker or where feedbacks cross stakeholder boundaries. Pataki et al. (2009) take the interaction
94
S.P. Shepherd
one step further and include the impact of forests. A multi-disciplinary study which used mediated
model building over a period of six months to build a model of the transport and ecosystem of
fossil fuel emissions is reported. The team conducted a whole ecosystem study of the role of
climate, urban expansion, urban form, transportation, and the urban forest in influencing net CO2
emissions in the Salt Lake Valley, Utah. The study showed the importance of the positive feedback
between urban developments and investment in transport infrastructure. Emissions were seen to
increase as this feedback created higher densities and increased traffic. The results suggested that
a strategy of doubling the density of tree planting would not have a significant impact on total
urban CO2 emissions, while land use and transport policies which combine to reduce urban sprawl
could produce a 22% reduction in CO2 emissions by 2030 compared with the business as usual
scenario.
Continuing with the theme of policies aimed at reducing environmental impacts, Han et al.
(2008) investigate emission reduction policies for the Inter-city transport problem. Sensitivity
analysis showed that reduction in CO2 is best achieved by increased development of the railway
network, but with a reduced emphasis on highway extensions and by imposing increases in fuel
taxes. Piattelli et al. (2002) study carbon taxes in the context of freight in Germany. Their model
covers investment in infrastructure, road, rail and waterway, subsidies, fuel costs and carbon taxes.
Carbon tax is shown to have only small impacts on mode share and raising tax on all fuels actually
increases the road share due to the relative costs of fuel by mode giving the counterintuitive result
that a negative carbon tax for freight movements would reduce road share. Carbon tax simply
increases revenues raised which could be used to reduce CO2 by other means but the study suggests
that investment in infrastructure for rail is better than a carbon tax, but even this has only small
impacts on mode shares.
Egilmez and Tatari (2012) also aim to reduce CO2 (in the USA), and three potential strategies
are tested with different levels of intervention: fuel efficiency, public transportation and electric
vehicle usage. Figure 8 gives an overview of the stock-flow diagram for their model showing
the links between population, GDP, wealth, vehicle stock, vehicle miles travelled, congestion,
emissions, highway capacity, and land use. Population and GDP growth rates are exogenous
while other equations are given in detail in this paper. The results showed that to be successful,
some hybrid of individual policies was crucial. All three policy approaches are required to meet
Figure 8. Stock-Flow diagram of the high-level model.

Source: Egilmez and Tatari (2012).
95
the CO2 targets in the USA 2 alone is not enough three are needed with at least two of them at
the 2nd level of implementation/success i.e. most optimistic. The authors also recognise the need
to change the objective from a traditional CBA approach to one which accounts for sustainability
if policies are to be adopted. This links nicely with a more general SD model of policy resistance
to change, Harich (2010); which discusses how social forces which favour change are inter-linked
with those which favour resistance to change and that a proper coupling is required to understand
why certain policies are ineffective.
European wide models have also been built to investigate EU wide transport and economic
developments. Fiorello, Fermi, and Bielanska (2010) provide an overview of the ASTRA model.
ASTRA links transport demand, the economy, with the vehicle fleet and environmental impacts
calculated for the European level. ASTRA has been used over the past decade to inform the
European Commission about the impacts of European transport policy especially over a long
time horizon, providing evidence for the White Paper on Transport. It is often used in conjunction
with other models also developed using a SD approach, such as the world energy model POLES
and the fleet development model TREMOVE. It is encouraging to note that whilst traditional
models such as the four-stage transport model still dominate the market in terms of research and
application areas that clients such as the EU are taking note of other approaches such as these
based on a SD perspective.
Finally within this section the policy of road pricing is studied by Shepherd (2013) who firstly
shows that a standard goal seeking structure from SD is equivalent to transport dis-equilibrium
modelling approaches in developing a simple model of competing toll operators. Shepherd then
shows how estimating response elasticities during periods of change or prior to a full equilibrium
can lead to errors in the toll setting strategies and cause instability and sub-optimal toll levels.
Liu, Triantis, and Sarangi (2010) develop a non-standard model of congestion pricing, combining
linguistic and fuzzy preferences plus social networking effects to model mode choice. Figure 9
shows part of the model covering peoples perception impacting on mode choice and the flow
Figure 9. Effect of peoples perceptions on the choice between bus and car.
Source: Liu, Triantis, and Sarangi (2010).
96
S.P. Shepherd
of revenues for reinvestment of toll revenue into the bus and Metro system which is modelled
over time. This demonstrates how SD can be used with other modelling approaches to provide a
different angle or viewpoint for a common policy issue.
7. Airlines and airports

Liehr et al. (2001) study the German airline market making use of data from Lufthansa. The aim
of the model was to identify cycle generating components of the airline industry. It is well known
that airlines aim to have high seat load factors so that they can maximise their profits. However
the study showed that due to long aircraft lead-times and delays in recognising when there is an
over-supply, the system begins to oscillate around the target seat load factor. Figure 10 shows the
basic structure of the model which contains two delays and a negative feedback loop which can
cause cycles even without external factors such as changes in GDP. Pricing strategies can only
dampen the effects and not remove the cycles. Leasing and quasicontinuous ordering policies
were shown to dampen out the impacts of cycles. The study compared traditional statistical and
static models with the SD model to improve acceptance of the model, but the dynamic model was
the only one which could describe why the business cycles occur and therefore was deemed to be
a better predictive model for out-performing the market when developing strategies for long-term
capacity and fleet planning.
Lyneis (2000) also used the passenger jet aircraft industry to demonstrate the benefits of SD
models over traditional trend extrapolation and regression models. The structure of the industry is embedded in the model and so allows the model to predict cycles in orders of aircraft
Figure 10. Basic structure of the airline market.

Source: Liehr (2001).
97
while the trend extrapolation approach overshoots the peak; and the regression model misses
it completely. Compared with Liehr et al., Lyneis model includes other reinforcing loops such
as competing airlines all attempting to gain market share during the upturns which then results
in overall capacity being overshot; the competition for new aircraft can lead to a longer lead
time which then leads to the need to project further ahead which all combine to amplify the
cycles.
Agusdinata and de Klein (2002) develop a causal loop analysis of airline alliances covering
the rush for globalisation during a transition economy, the impacts of congestion at hub airports,
uncertainty and distrust and the potential for collaborative learning. The analysis is linked to the
main literature in the air industry and the paper demonstrates how a CLD can be developed to
explain complex behaviour within a cyclic market. The paper whilst useful for discussing market
structure and responses, leaves the quantification of the model to future research.
Pierson and Sterman (2013) develop a model of airline industry profit cycles since de-regulation
in the US. The model expands the boundaries of previous work and contains four main elements,
endogenous capacity, demand, pricing and costs. The paper details the parameter estimation
process and the use of Markov Chain Monte Carlo methods to establish confidence intervals.
Contrary to prior work they show that the delay in aircraft acquisition is not as influential a
determinant of the profit cycle as others have found and that price setting strategies were found to
play a surprisingly important role in stabilising profits. These papers demonstrate the advantages
of SD approaches in that they are able to capture the cyclical market within the structure of the
model which provides more useful insights into strategies to deal with cycles as shown in Pierson
and Sterman (2013).
Investment in airport runway capacity has been studied with a simple model by Miller and
Clarke (2005, 2007). While both papers use the same model, the better description is in the later
paper. This paper develops and illustrates via a simple model a methodology for assessing the
strategic value of air transportation infrastructure, in particular the benefits of being able to react
swiftly to changes in the market. The model includes the influence of airport capacity on airline
Figure 11. Simple model of runway capacity and demand.

Source: Miller and Clarke (2007).
98
S.P. Shepherd
congestion costs which in turn affect the passenger fares and level of service, see Figure 11. These
factors then determine the aircraft per hour which in turn affect airport revenues. The model is
used to assess different strategies for infrastructure delivery. These strategies include: the level
of capacity increase, the time required to implement the capacity and the congestion threshold
which triggers the need for additional capacity. Monte Carlo simulation is used to account for
multiple sources of uncertainty. The model showed that a strategy of capacity enhancement based
on small increments and shorter response times could yield greater benefits than strategies that
consider larger capacity increases and which require longer response times. Future work could be
to link to an air market model. The earlier paper (Miller and Clarke 2005) combines real option
analysis via Monte-Carlo simulation with the model. The simulation is used to evaluate when the
option of building a second runway is greater than the real option of buying the land. In general
it shows that when an inflexible investment has an net present value (NPV) close to zero then the
flexibility to delay or advance a project can be most valuable. The issue of when to implement
large projects is something which is common in transport problems and SD models are a useful
tool in this area.
Suryani, Chou, and Chen (2010) extend this model to include impacts of population growth
and GDP and the impacts on both runway and terminal capacity. The model is used in a scenario
planning mode with optimistic and pessimistic cases. The difference in capacity requirements
suggested that the runway may cope with demand for an additional eight years in the pessimistic
case, and factors such as level of service requirements can have a significant impact on the floor
space required for the terminal. Suryani, Chou, and Chen (2012) adapt their model to investigate
air cargo demand and include effects of competition with other airports. This paper includes
a validation of the model against historic data along with a scenario analysis with high/low
growth scenarios to investigate when to invest in additional terminal capacity. A comparison with
statistical extrapolation approaches suggests the low growth scenario is more probable but the
authors suggest that the SD model is more useful as it includes the underlying structure of the
problem and hence the possibility to understand the implications of a more optimistic scenario
and the factors which influence the need for additional capacity.
Manataki and Zografos (2009, 2010) report on a generic planning tool for airport terminal
analysis. Whilst applied to a case study of Athens airport, the tool is designed to be generic
and easily applied to other airport terminal layouts. Their aim was to bridge the gap between a
very detailed operational analysis and a too aggregate macroscopic approach. The model splits
the terminal into a series of functional areas and facilities and is able to capture interactions
between these processes whilst allowing for random arrivals, delays, variations in schedule,
capacity and levels of service as well as basic design parameters. The approach uses a flight simulator or GUI to facilitate what if scenarios for the planner/decision-makers. Finally, Sgouridis,
Bonnefoy, and Hansman (2011) present the Global aviation industry dynamics (GAID) model
which includes three sectors, passengers, airlines and manufacturers. The study examines five
generic policies which aim to reduce the emissions of commercial aviation. Strategies investigated include, technological efficiency improvements, operational efficiency improvements, use
of alternative fuels, demand shift and carbon pricing or market-based incentives. Each policy
was implemented with three levels of effectiveness and combinations were tested and impacts
on CO2 , revenue passenger-kms, profitability and average fares were recorded. An interesting
aspect of the dynamics was the response to the highest level of carbon pricing where airline
profits suffer initially, but as airlines shed aircraft in response they are then able to set higher
fares and profits return to previous levels. This corresponds to real-world behaviour seen when
fuel prices were seen to double in 2008, initial disruption and some bankruptcies, followed
by mergers which effectively reduced capacities were observed. It seems that due to the long
timescales involved in the airline industry and in airport infrastructure coupled with changing
99
demands that SD has a lot to offer in this area and that more effort should be directed at such
issues.
8. Miscellaneous/emerging areas
As transportation covers a wide area, it has not been possible to categorise some papers within
the themes or areas identified above. This section includes some other application areas which
are very specific or which may point the way to new emerging areas for the application of SD in
the transportation field.
At the more detailed end of the transportation spectrum, Mehmood, Saccomanno, and Hellinga
(2003) apply SD to the area of car-following models. The proposed model assumed that drivers
were capable of estimating the spacing between their own vehicle and the next downstream vehicle
and rather than being based on drivers avoiding a collision, it is based on perception of safety. It
does not include the lead vehicles speed but does include status of brake lights. The model was
calibrated and validated against field data and proved to work well for three vehicles. However it is
not clear whether it would be practical to build a full micro-simulation model using this approach
and a more standard micro-simulation approach which takes on board the model suggested may
prove more useful in future work.
He et al. (2011) step up a level to investigate the problem of pollution around toll plazas in
China. The authors combine a computational fluid dynamics model with a SD model of toll plaza
operation to investigate the pollutants emitted under different assumptions about the penetration
rates and capacity (number of lanes) for electronic toll collection technologies. The SD part
was used to represent the queuing and service delays as vehicles go through the toll plaza. It
was demonstrated that there was some interaction between penetration rate and number of lanes
assigned to electronic toll collection which meant that pollutants may increase around the toll
booths as more lanes are provided under certain penetration rates.
Changing mode, Macmillan et al. (2014) used participatory group modelling to from a qualitative model of the major causal loops in the take up of cycling in Auckland, New Zealand. The loops
included a safety in numbers effect which reduces the real and perceived injury rate as cyclist
numbers increase. The model was populated with best available data and calibrated to recent
historic trends. Policies to encourage cycling were modelled over a 40-year period demonstrating
impacts on health, fuel savings and carbon emissions. It was meant to provide policy insight
rather than falsely precise predictions and sensitivity tests were used to check for any parameters
which either changed the ranking of policies or the order of magnitude of the outcomes. In this
way the model was useful for exploring policy and model assumptions and provided a basis for
discussions around cycling with others cities such as London.
Goh and Love (2012) develop two models to investigate policies to improve traffic safety.
The first model is used to assess policy options which aim to encourage the purchase of cars
with higher safety ratings. The second model is used to evaluate the impact of public transport
policies on travel time and traffic safety considerations. The first model looks at how cars with
different safety ratings are purchased and is similar to the work on alternative fuel uptake models,
network effects and price effects being included. The larger the current share the more attractive
it is to purchase due to lower maintenance costs and social norm factors. It was shown that
high taxes and subsidies for low-/high-rated cars are needed to overcome the network and price
effect. However, the sensitivity tests suggested that policies aimed at reducing the network effect,
e.g. awareness or early scrappage schemes for low-rated vehicles may improve the fleet average
star rating more quickly. The second model combined a goal-based safety policy which reacts
to a number of crashes versus a tolerable level but then added the link between this and the
policy on subsidies for public transport. It was shown that the larger the subsidy the greater the
100
S.P. Shepherd
reduction in crashes and hence less need for other safety interventions which could be more
expensive per saved crash. Similar to the work on the take up of AFVs, this paper demonstrates
the importance of long-term feedbacks and of taking a more holistic approach to the problem of
safety.
Bivona and Montemaggiore (2010) develop a case-study analysis of maintenance policies for
a city bus company. They describe the links between fleet maintenance, human resources and
training as well as other key policies such as fleet renewal and how these may impact on service
provision, customer satisfaction and overall profitability of the company. Two scenarios were
developed to try and prevent the decline in revenues. The first was to cut all departmental budgets
which while being effective in the short term were shown to be unsustainable in the long term.
This is a classic use of SD where performance measures are seen to get worse before then getting
better. The second policy included some investment in new buses, which reduced the average age
of the fleet, along with investment in training, all of which allowed increased service levels and
was shown to be more sustainable. This paper showed how the SD approach could be used with
decision makers to help them move away from simple cost-cutting policies to more sustainable
policies which at first sight appeared to be counter-intuitive.
Mayo, Callaghan, and Dalton (2001) report on a model for London Underground Limited
who understood that to evaluate different restructuring options would require a holistic and integrated system-wide view of the underground and its stakeholders. The model covered a range of
stakeholders including the market or customers who had a choice of mode, a workforce module including morale, experience, etc. suppliers, government and other private funding/finance.
Aspects such as capital assets and maintenance played an important role. Profit motivation and
investment levels could go into either virtuous or vicious circles depending on relative strengths
of feedback loops. The model was used to look at re-structuring options for LUL. Asset management, staffing continuity and collaboration early on were seen to be crucial in the success
of implementation of a public private partnership and hard to reverse. The model was provided
to the bidders to improve the tender process. The basic message to bidders from the model was
that an infrastructure company that maintains and invests appropriately in the asset base; maintains staff continuity; partners collaboratively with LUL and meets or exceeds the LUL-specified
performance targets will achieve significantly greater profits than one who does not. This paper
provides a good example of a model used in practice.
Randers and Gluke (2007) in a similar vein as the air market cycles, explain shipping markets
since 1950 as the interaction between two balancing feedback loops: a capacity adjustment loop
which creates a roughly 20-year wave, and a capacity utilisation adjustment loop which generates
a roughly 4-year cycle. They describe how they persisted with a small model rather than use a
detailed model and provided forecasts of when freight rates peaked. Shipping clients were at first
distrusting of the model but senior executives were convinced by the simpler models and some of
the client group trusted the forecasts to decide when to invest and when to get out of the market.
Another example of real-life impact and insight from a simple model which takes into account
stocks and flows.
Finally, Yeo, Pak, and Yang (2013) develop a model of the impacts of port security on the
competitiveness of the port. A CLD is developed with stakeholders and the quantitative model is
used to explore the impacts of increased security which on the one hand increases attractiveness
but on the other reduces it due to increased processing time and costs. Applied to a case study in
Korea, the study suggests that a more optimistic approach to security would increase trade and
prevent a spiral of decreasing competitiveness. Again the issue of feedback and policy resistance
is key to understanding the issues and in explaining how a more controversial or less favoured
policy may have the opposite impact to that expected by decision-makers whose mental model is
unable to take these factors into account.
101
9. A note on calibration and validation

Most SD models are not designed to provide point forecasts but rather they are built to display
the dynamic behaviour of the system under consideration. However being able to fit historic data
brings greater credibility to the model and a lot of effort is now put into behaviour reproduction
tests. Sterman (1984) reviews appropriate statistics for evaluating the historic fit of a model and
points out potential pitfalls of using some standard RMSE approaches. The use of Theils inequality
statistics which decomposes MSE into three components of bias, unequal variation and covariation
is recommended as it can be used to break down the sources of error. Fitting models to historic
data is though only a part of model validation for a system dynamicist. Barlas (1996) discusses the
model development and testing stages and emphasises other aspects relating to structure validity
tests. Sterman (2000) also argues that validation of a model includes a number of important tests
and that behaviour reproduction is only one part of the model validation process. Sterman (2000,
858) discusses 12 tests in detail including structure assessment, parameter assessment, extreme
condition tests, integration error and sensitivity analysis. The aim of the model building process is
more to learn about the dynamic behaviour of the system and to improve confidence in the model.
Despite this, the software platforms do now include automated calibration tools to aid the
estimation of parameters with the aim of improving historical fit. An example of this is described
in Pierson and Sterman (2013) described above where they make use of Markov Chain Monte
Carlo methods to establish confidence intervals. Other examples where the calibration process
is reported in detail and where models have been calibrated to historic trends and then validated
against data from a later period include those by Pfaffenbichler, Emberger, and Shepherd (2010),
and Haghani, Lee, and Byun (2003a).
10. Summary
In the 20 years since Abbas and Bell outlines the possibilities for the application of SD in transportation modelling, there has been a growing literature. Papers have appeared in some areas
more than others, as seen in the recent surge in the modelling of uptake of AFVs. This is in part
driven by the policy interest but is also an area which is well suited to the SD approach as it makes
use of existing structures such as the Bass diffusion model which has been applied to adoption
of new technologies in other areas. This structure is easily linked with choice models and fleet
ageing chains within the SD approach thus naturally accounting for time lags within the system.
As predicted by Abbas and Bell, strategic policy issues at a regional or national level involving
delays and feedbacks between different systems such as land use and transport have also been
developed as an area where SD has something to offer. Other areas which are well known for
their dynamics such as business cycles in the air market or for delays in responses as in the supply
chain area have also developed clusters of papers. The above review has outlined how the use of
both more qualitative CLDs and quantitative stock-flow models have been used to provide new
insights and explain the underlying structure within such systems. The approach often involves
stakeholders in the development of the CLDs and the simple nature of the stock-flow approach
aims at providing a transparent approach to modelling which is highly valued by stakeholders.
In the miscellaneous or emerging area section, it was seen that some applications were not
perhaps best tackled via SD. For example, SD is not appropriate for the traditional network
assignment problems, nor will it be able to replace micro-simulation tools. Whilst models can be
formed with spatial elements by making use of subscripts or arrays, from experience it seems that
once the number of zones increases beyond 200 then the model run times become a barrier and
large models are then run in compiled mode. Thus while large models are feasible in this mode,
it seems to go against the notion of a white box model and loses the benefits of communication to
102
S.P. Shepherd
stakeholders which comes with the use of smaller faster models which demonstrate the dynamic
behaviour of a system structure.
Instead the approach is better suited to providing a holistic system model which deals with
feedbacks and delays between actors in the system. Future applications could look at competition
dynamics, freight and the development of ports, sensitivity of systems and transport demand to
changing external factors related to demographics and the economy and in modelling behavioural
change whether this is at the user level of some higher level stakeholder. The approach allows
transport models to be easily linked to other sectors such as health, climate, and the economy
while taking into account time delays and feedbacks at different scales. However, it should be
noted that SD is not meant for precise point forecasts as Sterman (2002) summarises, System
dynamics helps us expand the boundaries of our mental models so that we become aware of and
take responsibility for the feedbacks created by our decisions.
System dynamics should be used to understand and explore the nature of the problem and gives
the modeller the opportunity to investigate general dynamic tendencies. The models can be used
to test which parameters play a significant role in the stability and response of the system and
the tools such as CLD and stock-flow diagrams enable a transparent approach to communicating
results with stakeholders including the use of flight simulators and gaming tools which other
approaches often lack. It is hoped that more transport modellers will in future work with the SD
approach and fulfil the aim of the international dynamics society which is to have a real-world
impact on business and policy. Finally, it is encouraging that clients such as the EU commission
and local authorities such as Leeds in the UK are commissioning SD based models to investigate
their specific problems, having seen the advantages of communication with their local elected
representatives.
Notes
1.
Rather than detail the various software platforms available the International system dynamics society
provides a list of key products at the following link: http://tools.systemdynamics.org/core-sd-software/
2. Bullwhip effect refers to a trend of larger and larger swings in inventory in response to changes in
customer demand, as one looks at firms further back in the supply chain for a product.
References
Abbas, K. A., and M. G. H. Bell. 1994. System Dynamics Applicability to Transportation Modeling.
Transportation Research Part A 28 (5): 373390.
Agusdinata, B., and W. de Klein. 2002. The Dynamics of Airline Alliances. Journal of Air Transport
Management 8 (4): 201211.
Barlas, Y. 1996. Formal Aspects of Model Validity and Validation in System Dynamics. System Dynamics
Review 12 (3): 183210.
Bivona, E., and G. B. Montemaggiore. 2010. Understanding Short- and Long-Term Implications of
Myopic Fleet Maintenance Policies: A System Dynamics Application to a City Bus Company.
System Dynamics Review 26 (3): 195215.
Chasey, A. D., J. M. de La Garza, and D. R. Drew. 2002. Using Simulation to Understand the Impact of
Deferred Maintenance. Computer-Aided Civil and Infrastructure Engineering 17 (4): 269279.
Disney, S. M., A. T. Potter, and B. M. Gardner. 2003. The Impact of Vendor Managed Inventory on Transport
Operations. Transportation Research Part E 39 (5): 363380.
Egilmez, G., and O. Tatari. 2012. A Dynamic Modeling Approach to Highway Sustainability: Strategies to
Reduce Overall Impact. Transportation Research Part A 46 (7): 10861096.
Fallah-Fini, S., H. Rahmandad, K. Triantis, and J. M. de la Garza. 2010. Optimizing Highway Maintenance
Operations: Dynamic Considerations. System Dynamics Review 26 (3): 216238.
Feng, C. M., and C. H. Hsieh. 2009. Effect of Resource Allocation Policies on Urban Transport Diversity.
Computer-Aided Civil and Infrastructure Engineering 24 (7): 525533.
103
Fiorello, D., F. Fermi, and D. Bielanska. 2010. The ASTRA Model for Strategic Assessment of Transport
Policies. System Dynamics Review 26 (3): 283290.
Flynn, P. 2002. Commercializing An Alternate Vehicle Fuel: Lessons Learned From Natural Gas for
Vehicles. Energy Policy 30 (7): 613619.
Forrester, J. 1958. Industrial Dynamics A Major Breakthrough for Decision Makers. Harvard Business
Review 35 (4): 3766.
Friedman, S. 2006. Is Counter-Productive Policy Creating Serious Consequences? The Case of Highway
Maintenance. System Dynamics Review 22 (4): 371394.
Georgiadis, P., D. Vlachos, and E. Iakovou. 2005. A System Dynamics Modeling Framework for
the Strategic Supply Chain Management of Food Chains. Journal of Food Engineering 70 (3):
351364.
Ghaffarzadegan, N., J. Lyneis, and G. P. Richardson. 2011. How Small System Dynamics Models Can Help
the Public Policy Process. System Dynamic Review 27 (1): 2244.
Goh, Y. M., and P. E. D. Love. 2012. Methodological Application of System Dynamics for Evaluating
Traffic Safety Policy. Safety Science 50 (7): 15941605.
Haghani, A., S. Y. Lee, and J. H. Byun. 2003a. A System Dynamics Approach to Land Use / Transportation
System Performance Modeling. Part 1: Methodology. Journal of Advanced Transportation 37 (I):
141.
Haghani, A., S. Y. Lee, and J. H. Byun. 2003b. A System Dynamics Approach to Land Use / Transportation
System Performance Modeling. Part 2: Application. Journal of Advanced Transportation 37 (I): 4382.
Han, J., and Y. Hayashi. 2008. A System Dynamics Model of CO2 Mitigation in Chinas Inter-City
Passenger Transport. Transportation Research Part D 13 (5): 298305.
Hang, W., and X. Li. 2011. Application of System Dynamics for Evaluating Truck Weight Regulations.
Transport Policy 17 (4): 240250.
Harich, J. 2010. Change Resistance as the Crux of the Environmental Sustainability Problem. System
Dynamics Review 26 (1): 3572.
Harrison, G., and S. P. Shepherd. 2014. An Interdisciplinary Study to Explore Impacts From Policies for
the Introduction of Low Carbon Vehicles. Transportation Planning and Technology 37 (1): 98117.
He, J., Z. Qi, W. Hang, M. King, and C. Zhao. 2011. Numerical Evaluation of Pollutant Dispersion at a
Toll Plaza Based on System Dynamics and Computational Fluid Dynamics Models. Transportation
Research Part C 19 (3): 510520.
Janssen, A., S. F. Lienin, F. Gassmann, and A. Wokaun. 2006. Model Aided Policy Development for the
Market Penetration of Natural Gas Vehicles in Switzerland. Transportation Research Part A: Policy
and Practise 40 (4): 316333.
Kohler, J., M. Wietschel, L. Whitmarsh, and W. Schade. 2010. Infrastructure Investment for a Transition to
Hydrogen Auotomobiles. Technological Forecasting & Social Change 77 (8): 12371248.
Kwon, T. 2012. Strategic Niche Management of Alternative Fuel Vehicles: A System Dynamics Model of
the Policy Effect. Technological Forecasting & Social Change 79 (9): 16721680.
Liehr, M., A. Groler, M. Kleinb, and P. M. Millinga. 2001. Cycles in the Sky: Understanding and Managing
Business Cycles in the Airline Market. System Dynamics Review 17 (4): 311332.
Liu, S., K. P. Triantis, and S. Sarangi. 2010. A Framework for Evaluating the Dynamic Impacts of a
Congestion Pricing Policy for a Transportation Socioeconomic System. Transportation Research Part
A 44 (8): 596608.
Lyneis, J. M. 2000. System Dynamics for Market Forecasting and Structural Analysis. System Dynamics
Review 16 (1): 325.
Macmillan, A., J. Connor, K. Witten, A. Kearns, D. Rees, and A. Woodward. 2014. The Societal Costs and
Benefits of Commuter Bicycling: Simulating the Effects of Specific Policies Using System Dynamics
Modeling. Environmental Health Perspectives 122 (4): 335344.
Manataki, I. E., and K. G. Zografos. 2009. A Generic System Dynamics Based Tool for Airport Terminal
Performance Analysis. Transportation Research Part C 17 (4): 428443.
Manataki, I. E., and K. G. Zografos. 2010. Assessing Airport Terminal Performance Using a System
Dynamics Model. Journal of Air Transport Management 16 (2): 8693.
Mayo, D. D., M. J. Callaghan, and W. J. Dalton. 2001. Aiming for Restructuring Success at London
Underground. System Dynamics Review 17 (3): 261289.
Mehmood, A., F. Saccomanno, and B. Hellinga. 2003. Application of System Dynamics in Car-Following
Models. Journal of Transportation Engineering 129 (6): 625634.
Meyer, P. E., and J. Winebrake. 2009. Modeling Technology Diffusion of Complementary Goods: The Case
of Hydrogen Vehicles and Refueling Infrastructure. Technovation 29 (2): 7791.
104
S.P. Shepherd
Miller, B., and J. P. Clarke. 2005. Investments under Uncertainty in Air Transportation: A Real Options
Perspective. Journal of the Transportation Research Forum 44 (1): 6174.
Miller, B., and J. P. Clarke. 2007. The Hidden Value of Air Transportation Infrastructure. Technological
Forecasting & Social Change 74 (1): 1835.
Otto, A., and H. Kotzab. 2003. Does Supply Chain Management Really Pay? Six Perspectives to Measure
the Performance of Managing a Supply Chain. European Journal of Operational Research 144 (2):
306320.
Park, S. Y., J. W. Kim, and D. H. Lee. 2011. Development of a Market Penetration Forecasting Model for
Hydrogen Fuel Cell Vehicles Considering Infrastructure and Cost Reduction Effects. Energy Policy
39 (6): 33073315.
Pataki, D. E., P. C. Emmi, C. B. Forster, J. I. Mills, E. R. Pardyjak, T. R. Peterson, J. D. Thompson, and E.
Dudley-Murphy. 2009. An Integrated Approach to Improving Fossil Fuel Emissions Scenarios with
Urban Ecosystem Studies. Ecological Complexity 6 (1): 114.
Pfaffenbichler, P. 2011. Modelling with Systems Dynamics as a Method to Bridge the Gap between Politics,
Planning and Science? Lessons Learnt from the Development of the Land Use and Transport Model
MARS. Transport Reviews: A Transnational Trans-disciplinary Journal 31 (2): 267289.
Pfaffenbichler, P., G. Emberger, and S. P. Shepherd. 2010. A System Dynamics Approach to Land Use
Transport Interaction Modelling: The Strategic Model MARS and Its Application. System Dynamics
Review 26 (3): 262282.
Pfaffenbichler, P., R. Krutak, and S. Renner. 2011. Modelling the Development of Vehicle Fleets with
Alternative Propulsion Technologies. ECEEE 2011 summer study, Belambra Presqule de Giens,
France.
Piattelli, M. L., M. A. Cuneo, N. P. Bianchi, and G. Soncin. 2002. The Control of Goods Transportation
Growth by Modal Share Re-planning: The Role of a Carbon Tax. System Dynamics Review 18 (1):
4769.
Pierson, K., and J. D. Sterman. 2013. Cyclical Dynamics of Airline Industry Earnings. System Dynamics
Review 29 (3): 129156.
Potter, A., and C. Lalwani. 2008. Investigating the Impact of Demand Amplification on Freight Transport.
Transportation Research Part E 44 (5): 835846.
Randers, J., and U. Gluke. 2007. Forecasting Turning Points in Shipping Freight Rates: Lessons From 30
Years of Practical Effort. System Dynamics Review 23 (2/3): 253284.
Sgouridis, S., P. A. Bonnefoy, and R. J. Hansman. 2011. Air Transportation in a Carbon Constrained World:
Long-Term Dynamics of Policies and Strategies for Mitigating the Carbon Footprint of Commercial
Aviation. Transportation Research Part A 45 (10): 10771091.
Shen, Q., Q. Chen, B. Tang, S. Yeung, Y. Hu, and G. Cheung. 2009. A System Dynamics Model for the
Sustainable Land Use Planning and Development. Habitat International 33 (1): 1525.
Shepherd, S. P. 2013. Toll Competition and Dynamic Toll Setting Strategies. International Journal of
Sustainable Transportation 7 (3): 186203.
Shepherd, S. P., P. W. Bonsall, and G. Harrison. 2012. FactorsAffecting Future Demand for ElectricVehicles:
A Model Based Study. Transport Policy 20 (March): 6274. doi:10.1016/j.tranpol.2011.12.006
Stepp, M. D., J. J. Winebrake, J. S. Hawker, and S. J. Skerlos. 2009. Greenhouse Gas Mitigation Policies
and the Transportation Sector: The Role of Feedback Effects on Policy Effectiveness. Energy Policy
37 (7): 27742787.
Sterman, J. D. 1984. Appropriate Summary Statistics for Evaluating the Historical Fit of System Dynamics
Models. Dynamica 10 (2): 5166.
Sterman, J. 2000. Business Dynamics: Systems Thinking and Modeling for a Complex World. Boston:
Irwin/McGraw-Hill.
Sterman, J. D. 2002. All Models are Wrong: Reflections on Becoming a Systems Scientist. System
Dynamics Review 18 (4): 501531.
Struben, J., and J. D. Sterman. 2008. Transition Challenges for Alternative Fuel Vehicle and Transportation
Systems. Environment and Planning B-Planning & Design 35 (6): 10701097. www.envplan.com
Suryani, E., S. Y. Chou, and C. H. Chen. 2010. Air Passenger Demand Forecasting and Passenger Terminal Capacity Expansion: A System Dynamics Framework. Expert Systems with Applications 37 (3):
23242339.
Suryani, E., S. Y. Chou, and C. H. Chen. 2012. Dynamic Simulation Model of Air Cargo Demand Forecast
and Terminal Capacity Planning. Simulation Modelling Practice and Theory 28 (November): 2741.
Tako, A., and S. Robinson. 2012. The Application of Discrete Event Simulation and System Dynamics in
the Logistics and Supply Chain Context. Decision Support Systems 52 (4): 802815.
105
Trappey, A. J. C., C. Trappey, C. T. Hsiao, J. J. R. Ou, S. J. Li, and K. W. P. Chen. 2012. An Evaluation Model
for Low Carbon Island Policy: The Case of Taiwans Green Transportation Policy. Energy Policy 45
(June): 510515.
Ulli-Beer, S., F. Gassmann, M. Bosshardt, and A. Wokaun. 2010. Generic Structure to Simulate Acceptance
Dynamics. System Dynamics Review 26 (2): 89116.
Walther, G., J. Wansart, K. Kieckhafer, E. Schnieder, and T. S. Spengler. 2010. Impact Assessment in
the Automotive Industry Mandatory Market Introduction of Alternative Powertrain Technologies.
System Dynamics Review 26 (3): 239261.
Wang, J., H. Lu, and H. Peng. 2008. System Dynamics Model of Urban Transportation System and Its
Application. Journal of Transportation Systems Engineering and Information Technology 8 (3): 8389.
Wilson, M. C. 2007. The Impact of Transportation Disruptions on Supply Chain Performance. Transportation Research Part E 43 (4): 295320.
Wolstenholme, E. F. 2003. Towards the Definition and Use of a Core Set of Archetypal Structures in System
Dynamics. System Dynamics Review 19 (1): 726.
Xu, H., A. N. Mashayekhib, and K. Saeed. 1998. Effectiveness of Infrastructure Service Delivery Through
Earmarking: The Case of Highway Construction in China. System Dynamics Review 14 (2): 221255.
Yeo, G. T., J. Y. Pak, and Z. Yang. 2013. Analysis of Dynamic Effects on Seaports Adopting Port Security
Policy. Transportation Research Part A 49 (March): 285301.

Rseg

Загружено:

Сведения о документе

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Rseg

Загружено:

Авторское право:

Доступные форматы

A review on integrated active steering and braking control for vehicle yaw stability system.

on Integrated Active Steering and Braking Control for Vehicle Yaw

Received 1 January 2014

2014Penerbit UTM Press. All rights reserved.

the driver to keep the vehicle stable on the desired path. By

71:2 (2014) 105111 | www.jurnalteknologi.utm.my | eISSN 21803722 |

2.0 VEHICLE MODELS

I w i Rwi Fxi Tei Tbi

where i is wheel angular acceleration, Rwi is wheel radius, I wi

2.1 Nonlinear 7 DOF Vehicle Model

Vehicle motion with nonlinear tire forces represents a nonlinear

Fx3 Fx 4 ( Fy1 Fy 2 ) sin f

d Fx1 cos f Fx 2 cos f

2 Fy1 sin f Fy 2 sin f Fx3 Fx 4

Table 1 Vehicle parameters

2.2 Linear Bicycle Model

In vehicle dynamics studies, each wheel represents 1 DOF. Thus,

The classical linear bicycle model as shown in Figure 2 is the

and k us is understeer parameter as given in equation (8). Notice

3.0 YAW CONTROLOBJECTIVES& PROBLEMS

sideslip angle and yaw rate r . Notice that vehicle speed v is

Figure 2 Bicycle model

2.3 Reference Model

where k us is known as understeer parameter and defined as

In vehicle yaw stability control system, the variables of yaw rate r

M z as implemented in [2, 3, 5, 10, 11, 16, 17, 27, 29, 30].

INTEGRATED ACTIVE STEERING & BRAKING

In vehicle yaw stability system, yaw rate and sideslip tracking

In recent years, an integration of active front steering (AFS) and

Figure 3Control structure of integrated AFS-DYC control

To overcome the presence of uncertainties and crosswind

Studies in [1, 4,22] considered the road adhesion coefficient,

uncertainty of tire cornering stiffness during variations of driving

HIGH PERFORMANCE ROBUST TRACKING

In previous sections, the control algorithms to ensure a robustyaw

As in the above studies, the use of CNF control could

Figure 4Transient response improvement using a CNF control technique

system, fuzzy and neural network system, electrical motor and

The nonlinear function (r , y) is used to change the

with e1 (t ) and e2 (t ) representingthe yaw rate and sideslip

The authors are grateful to the Ministry of Education of Malaysia,

S.C. Baslamisli, I. E. Kse, G. Anla.2009.Gain-scheduled integrated

[23] J. Wang, R. G. Longoria. 2009.Coordinated and reconfigurable vehicle

[47] Bijnan Bandyopadhyay, Kyung-Soo Kim, Fulwani Deepak.2009. Sliding

2015 International Conference on Unmanned Aircraft Systems (ICUAS)

A Review on the Platform Design, Dynamic Modeling and Control of

the miniature hybrid UAVs to become more mature, in terms

Examples of Manned Hybrid Aircraft.

II. P LATFORM D ESIGN

utilize any coaxial rotor because the differential thrust can

Examples of Tilt-Rotor UAVs.

operate only during vertical flight to provide extra lift for

of the projects was successful [4][30]. Tail-sitters can be

Examples of Convertiplane UAVs.

of the Rotor-Wing works show similar design configuration

key components are contained. Detailed explanations of these

Examples of Tail-sitter UAVs.

Representative Modeling Work

[15], [19], [20], [50], [51]

provide larger wing span. Moreover, the wings retract during

highlighting the particular reason for their selection.

systems (such as Nacelle axis system, hub-axis system, and

B. Representative Modeling Work

experiments. Furthermore, [54] also addresses a technique of