Proceedings of the 2000 IEEE

International Conference on Robotics & Automation

San Francisco, CA • April 2000

Design of Steering Mechanism and Control

of Nonholonomic Trailer Systems
Yoshihiko Nakamura *l Hideaki Ezaki .1 Yuegang Tan .2 Woojin Chung *a

Dept. of Mechano-Informatics *l Mechanical and Electrical Engineering .2 Robotics Research Center *a

University of Tokyo Wuhan University of Technology Korea Institute of Science and Technology
7-3-1, Hongo, Bunkyo-ku 122 Luoshi Road 39-1, Hawolkok-Dong, Sungbuk-ku
Tokyo 113-8656, .lapin Wuhan, P.r. China Seoul 136-791, Korea
[nakamura,hide]¢Oynl.t.u-tokyo.ac.jp tanzhang~pu blic.wh.h b.cn wj('hu ng(@kistmail.kist.re.kr

Abstract one. Being converted to the chained form, a tractor

A wheeled mobile robot with trailers has been. studied
as a class of nonholonomic systems. It is proved that
a system of a tractor and trailers with an app,vpriatc
connecting "mechan,ism can be stabilized to desired po-
sitions via nonbolonomic motion control. Trailers, on
the other hand, have bee,t developed and widely used
in lhe industry. The m.ain focus of industrial design
is set on reducing tracking error from a reference tra-
jectory, This paper atlempls to bridge over the gap
between these lwo approaches. We develop a design
theory of trailer systems with passive steering. The
designed systems show a good performance in practical
path followi.ng, and accept the chained form transfor-
mation and nonlinear control st~ttegies for nonholo-
heroic systems.
Keywords: Nonholonomic constraints, Trailer sys-
tern, Passive steering, Path following, Practical mech-
allisln
1 Introduction
Nonholonomic mechanical systems have received
much attention in tile fields of robolics and control.
Laumond[1] proved hy construction that a car-like
robot is controllable even wii,h the limited steering
angle. Laulnond and Sim6on extended the heuristic
approach to a mobile robot with one l.railer[2]. Theo-
retic approaches, then, followed in control commmfity
based on the differential geometric framework. Mur-
ray and Sastry[3] named a simple nonlim'a.r system
the chained form and proposed to use it. as a canon-
ical form for a class of drifth'ss syst.ems. S0rda.len[4t]
showed a mobile robots with n trailers to be convert-
ible t.o the chained form, where a trailer is assumed
a passive two-wheeled car without steering and being
connected to the previous trailer/tra, ctor with a king-
pin hitch at the center of wheel shaft of the previous
and all trailers can be stabilized to desired positions
via. esla.I)lished nonholonomie motion control strate-
gies for the chained form[5, 6, 7]. S0rdalen, Naka~
mura and Chung[8] developed "t,he nonholonomic ma-
nipulator," whose kinematic model is similar to those
of trailer systems. The real-world tra~ject.ory gener-
ated by the nonholonomic motion control tends to
contain high frequency oscillations or large overshoot,
although the mathematical trajectory in the chained
form enjoys a nice convergent profile. Chung and
Nakamura[9] then proposed to consider mathemati-
cal simplicity in the chained-form transformation and
designed "the chained form manipulator."
On tile other hand, trailers have been developed
and widely used in the industry. The main focus of
such design was on reducing passive tracking error
fi'om t,ractor's trajectory, and little attention was paid
on active motion cent.tel.
Automated trailer systems will find an important
part in automated factories, ports or airports, if the
technology is available. Nonholonomic motion plan-
ning/control will be applied when loading and unload-
tug freighl..
In this paper, we all.erupt to fidfill the gap be-
tween t.he nonlinear control theory for nonholonomic
trailer systems and the mechanical design for good
passive tracking performance of indust.rial trailer sys-
tems. Based on the knowledge of clm.ined form trans-
formation, we design a new passive steering mech-
a.nism and show that it, provides a trailer with the
both properties; namely, the chained-form convertihil-
ity and a good passive tracking performance. A pro-
retype is fabricated and used t.o verify the proposed
design theory.

0-7803-5886-4/00/$10.00© 2000 IEEE 247

2 Kinematic Models of Trailer System 1
Oi = "-7(-Gi sin ¢i + sin(0i_l - Oi + ¢i--1) )Vi--1 (6)
al
with Steering
i
2.1 The Kinematic Model vi = H Gjvo (7)
j=l
We will compare performances of different mechan- (i= a,...,,0
ical design of trailers with and without, steering. The
most general design of trailer system is shown in Fig. where
1, from which all the comparative designs we discuss Gi dgr cos(0i_l -- Oi + 0i-1)
are derived by appropriate choice of parameters. (s)
COS 0i
Tractor Note that the above model is valid ill a limited range
of
7r 71-
--g<¢~<~ (i=l,...,n). (9)
When ¢,: = 0(i = l , . - . , n), this model is equivalent t.o
the Sordalen's trailer niodel without st.eering[4].
The above model also tel)resents four wheel trailer
///d. TraderI n systems, which is rather common in industry, with
setting lhe angle of Dent and rear axle of ]c~h trailer
Figure 1: Trailer system with steering. as 02~.-L and 0,,~:,
Tire linear and angular velocity of the tractor are
2.2 T r a i l e r s in t h e I n d u s t r y
represented by v0 and w, and considered the inputs.
The i °~ connecting rod is fixed to the frame of i th 2.2.1 Yalnamiya Model
trailer and is di(i = 1 , . . . , n ) in length and Oi(i =
1,...,.n) in orientation. The intersection of the i th We discuss two trailer systems with passive steering in
and i + 1th rods is called the center of tile i th trailer, of the industry. The first one is found in a recent patent
which linear velocity is represented by vi(i = 1 , - . - , n). file in Japan[10]. The Yamamiya model is shown in
The steering angle of tile wheels, ~)i(i = 1 , . - - , n), is Fig. 3, where steering angles are determined by rods A
defined with respect, to the rod of the i th trailer, and front..
¢0 is set to zero.
Tile geometric relations between the variables is • ° ......

shown in Fig. 2 and formatted by

i-1 ]y/~ 0 vertical axis 0 ~ e e to rotate) • coupler
(free to rotate)
Figure 3: Yamanfiya's trailer.
through F. Since rods A and B are fixed, front wheels
are steered in the same direction as that of A. Rear
~ o Traikr i-I ~"~"-0 i
wheels are oriented by rods D through F ill tire oppo-
Trailer i
Figure 2: Geometric relations of velocity. site direction to the fl'ont ones.
'Fhe Yamanfiya model can bo represented in (he
form of Eqs.(3) (8), by setting an alq)roximation on
I'iCOSQi =-- t'i_l COS(Oi_l -- Oi @0i-1) (1)
the connecting point of rod as shown ill Fig. 4. Now,
1
Oi= ~ ( - - v i s i n O i + v i - l s i l l ( O f ~ - - O i +oi-j)) (2)

The kinematic model after all is rel)resented in recur- Before approximation After approximation
sion and sunnnarized by the following equations. Figure 4: Approxinmtion of Yanlanliya's trailer.
~i in 1~;(ts.(3)-(8) is w r i t ( e l l as
.i: = cos 0,, v,, (3)
= sin 0,, v,,. (4) 0 (i : odd)
(10)
0'0 = w (,5) @ : 0i - 0~-1 (i:evc~)

248
2.2.2 ITS Model
The ITS (Intelligent Transport System) projects have
been studied in the U.S. and Europe as well as in
J a p a n to reduce traffic j a m and accidents by making
full use of the latest information technology. At truck
platoon system was conducted in the C H A U F F E U R
Project. Each driverless truck is steered through
car-to-car communication[11], which is regarded as a
trailer system connected by invisible rods. The pro-
posed basic PD control law is,
6i = k1,(di)pi + kd(di)tii
where tti is the relative angle and d i is the distance
between the i th and i - I th trucks. ~5i is tile steering
angle of the i th truck, kz,(di ) is determined such that
the rear wheels of tile i th truck precisely follow those
of the i - 1th truck when they move on circular trajec-
tories, k d ( d i ) is called the dynamic stabilizer and s e t
zero in this paper.
Rewriting Eq.(11) yields the I T S model,
0.5(0i+l-0i) (i:odd)
¢i = 0 (i : e v e n )
3 The Three-Point Model
We introduce the three-point model, the new steer-
ing method for trailer systems. The three-point model
is named after the property that the steering angle of
each trailer is determined by three points, namely, the
center points of the previous, current, and following
trailers.
Steering angle, shoukl be controlled in practice by
simple link mechanisms, instead of electrical devices.
Simplicity of tile three-point model satisfies this re-
quest. We propose concrete mechanism designs in Sec-
tion 6.
'lb the kinelnatic model of Eqs.(3) through (8), the
three-l)oint model adds the following:
61=r(O,+l-Oi) (i=l ..,n.-l)
4~,, = 0
where r is called tile steering gain. Note that the last
trailer is not steered.
4 Design O p t i m i z a t i o n in the Three-
Point M o d e l
4.1 Optimal Steering Gain
In this subsection the optimal steering gain is to be
discussed by nmnerical]y investigating the relationship
between r and the path following performance. Sim-
ulations are done for the 23 dimensional model, of a
tractor with 20 trailers. The index of path follow-
ing error is the mean of time average of trailer errors.
Each trailer's error is the normal distances from the
reference trajectory to the center of trailer. Three typ-
ical trajectories of a tractor are defined and shown in
Table 1, where di = 1(i = 1 , . . . , n), R and L are the
radius and the length of clothoidal curve.
Table 1: Tractor's trajectory
clothoid R L = 100
(11) circle x2 + y2 = 5 2
sinusoid y = 2(1 - cos(0.7x))
'File results of simulation are shown in Figs. 5
through 7. Note that the case with r = 0 corresponds
to Sordalen's model. The temporal change of a trailer
system is shown by solid lines, among of which the cir-
cled ones indicate the initial and final configurations.
A dotte(l curve shows the trajectory of the tractor.
The n th trailer is initially located at the origin. Fig-
ures 5, 6 and 7 show the results for clothoidal, circular,
and sinusoidal trajectories, respectively.
(12)
r=--I r = 0 (Sordalen's model)
-1
r=0.5 r = 0.53
(13) Figure 5: Motion of trailer systeni with clothoidal tra-
(14) jectory.
From t,hese simulation results and those carried out
with different va.lues of path p a r a m e l e r s , a single im-
port.ant feature is obserw, d. When I' is less than 0.5,
lhe profile of trailer sysi.em tends to move inner and
inner or shrink as time goes on. On the other hand if I'
grows even slightly over 0.5, the trailer profile starts t.o
show a zigzag motion. Repeatedly, the fact was com-
mon for all of clothoidal, circular, and sinusoidal paths
and for many other parameter choices of the paths.
249
 in the practical path following through the numerical
analysis of relationship between the steering gain and
the path following performance.
The authors understand the importance of analyt-
ical investigation, although the numerical sinmlation
was adopted in the analysis in this section. The index
r=-I = 0 (Sordalen's model' of performance, unfortunately, shows high nonlinear-
ity including path parameters dependent on the choice
of path as well as the configuration variables that
evolve nonintegrable differentiM equation, It is left for
the fl~ture research to provide the optimal three-point
model with an analytical insight. In particular, it is
interesting l,hat the emergent of zigzag trailer profile
is vary sensitive wilh the growth of r beyond 0.5.
r=0.5 r = 0.51 4.2 Comparison with Industry Designs
Figure 6: Motion of trailer system with circular tra- We compare path following performance of the
jectory. Sordalen model, the optimal three-point model, the
Yamamiya model and lhe ITS mo(lel. Figures 8
through 10 show the behaviors of the Yamamiya and
,0 ITS models. Behaviors of tile Sordalen model and
optimal three-point model are seen in right-top and
left-bottom of Figs. 5 through 7, respectively. Table 2
shows the path following error in summary. The error
i
is measured based on the same definition described in
i ~0 ,'s ia ~is 1o is g w is N is J0 ,is the previous subsection, and normalized by that of the
r= --1 r = 0 (S¢,rda.len's model) optimal three-point model.

rraa~
+J ' x . o f N ~ . / ' V ~
M~rmr -, I,r~J

a
o ~ Io ~s ~n #5 ~o is
r=O.5 r : 0.515

Figure 7: Motion of trailer system with sinusoidal t,ra,- Yamamlya model ITS model
jectory. Figure 8: Motion of trailer system with clothoidM tra-
jectory.

We, therefore, wouhl like to call a. three-point model

with r = 0.5, the oplimol thr~-poinl mo&l. Oi)timal
three-point model orients the i tt' tra iler's wheel to the
langent of a virtual circle which passes through the
centers of the i - 1°', F h, and i + i th lra ilers. Some
mobile robols in the lileral, ure, including the snake-
like vehicle [12], have the same steering mechanisnl
as lhe ()ptimal Ihree-l~oint model. To tile besl of the
authors' knowledge, there is no industrial design of Yamamiya model ITS m(Mel
trailer systerns adopted the optimal t hree-poinl model, Figure 9: Motion of trailer systenl with circular tra-
unfortunately. It is because the advantage of such de- jectory.
sign had neither been well investigated nor been suf-
ficiently stressed. A contribution of this paper is 1o The opt, ilnal three-point model shows the best per-
hilight tile advantage of the optimal |hree-point model formance as far as these trajectories are considered,

250
 follows: Substituting Eq.(13) in Eq.(6) and replacing
lO ,d
vi-1 with - v i _ 1 yields
1
rrm~r n 0i = - ~ i {cos(1 - r)(Oi_, - Oi)tan r(0i - 0i+1)
-s ~al --> r~l

-la 10
+ s i n ( 1 - r)(Oi_] - Oi) }vi-] (15)
Q ~ t0 is
x
~0 ~s 5o ' 0 g w i~
1
k0 ~s ~ ~s 1
Yamamiya model ITS model = - ~-,.{sin r(Oi - 0i+]) (16)
Figure 10: Motion of trailer system with sinusoidal + cos r(Oi - Oi+~)tan(1 - r)(0i-] - Oi)}vi
trajectory.
We then reverse indexing of a tractor and trailers by
Table 2: Error from reference trajectory
replacing Oi+a wil,h 0i-1, and vi with vi-1 in Eq.(16).
Sordalen Y a m a m i y a l ITS
Then,
elothoid 20.11 10.85 i 2.29
circle 10.75 4.17 ii 1.66 , 1
= g{cos,-(0f - rant1 - ,.)(0,: -
sinusoid 2.43 1.94 , 2.12 i
+ s i n r(0;_l - 0i))}vi-i (17)
The behavior of these models, in general, resemble the
three-point model with r < 0.5. The reader should notice that the above equation is
equivalent to Eq.(6) with 0i = (1 - r)(Oi+a - Oi). In
4.3 Backward Motion other words a backward motion of the system with
sleering gain r is equivalent to a forward motion of the
Backward motion of trailer systems has not been
system with steering gain 1 - r. Therefore, if and only
discussed in the literature, which seems due to the
if r = 0.5, a trailer system shows non-zigzag motions
fact that it apparently tends to show a zigzag or un-
in both forward and backward directions.
stable motion. Figure 11 shows the results of simula.-
lion when the tractor drives trailers backward along Since the fi-ont trailer in the backward motion has
the dotted trajectory. fixed steering On --- 0, the trailer system has to fol-
lowed the tangential straight line. Therefore, if we
provide the n ~h trailer with an active steering, the new
trailer system can be completely driven backward di-
rection as well as in the forward direction. In the for-
ward motion, we fix ¢,, to 0. In the backward motion,
the angular velocity of the n th trailer is driven by the
n th trailer's on-board actuator, while the linear veloc-
ity of the n ta trailer is driven by the tractor's push-
ing force. The tractor is controlled so as t.o keel) the
S¢,rdalen model Optimal three-poin! mode] wheel shaft crossing perpendicular to the first, trailer's
rod. The pushing velocity v0 is calculated from v,,. via

i Eq.(7).
5 Differential Kinematic Analysis and
Control of the Three-Point Model
5.1 Chained Form Convertibility
The chained form[3] is a canonical form for a
Yammniya model ITS model class of driftless nonholonomic systems. Thanks to
Figure 11: Backward motion of trailer system with the simplicity of the chained form, many usefid con-
circular trajectory. trol schemes have been devek)ped. 'Fhe sufficient
conditions[7] For converting of a kinematic model to
Except for the optimal three-poinli model, all the the chained form are,
trailer models showed the zigzag motion. The opti- (A) Kinematic model being convertible into a trian-
mal three-point mode/ resulted in stable motion, but gular structm'e.
followed a straight line that is tangentially connected (B) The triangular structure satisfies Eq. (43) in ap-
to the circular reference. The reason is explained as l)endix A.

251
Equations (3)-(8) are rewritten as, Of.+2 1 - r
- - - (30)
0aa+l d, cos x,~+~ cos2(1 - r ) ( x , + l - x,~+2)
y = v tan0,, (18)
0f,,+a 1 (31)
1 (~Xn+ 2 COS2 J2n+ 2
Oi = ~ 7 ( - G i sill 0 i + s i n ( 0 i - I -- Oi + ¢ i - 1 ))
i

n (i = l , - . . , n ) (19) In the range of Eq.(9),

cosOnI-Ij=iG j

def ,
COSr(Xi -- X i + l ) :~ 0 (i = 3,-.., n + 1) (32)
where, v = x. If 0i = (}i(Oi-1), we call converl the
kinematic model into an n + 3 dimensional triangular Therefore, condition (B) is satisfied if v # 1. This
structure by setting concludes that tile three point model is chained-form
convertible. Scrdalen and ITS nlodels also satisfy tile
(20) chained-form convertibility, hut Yamanliya nlodel or
most conlmon design of trailer syst(,ms ill industry do
[,, .., 0,,, (21)
not enjoy this property.
Tile tllree point model sat, isfies
5.2 Control for the Chained Form
0i=0i(0~) (i=l,. ..-1) (22) We will show, for exanlple, an open-loop control
0,, = 0.(0,,-~) (23) with time polynonlials[13]. The inl)ut of the chained
form is written by
Next., f i in Eq. (40) are expressed as,
vl = b0 (33)
tall r(xa - x4) + tan(a~2 - x3)
f3= d l cos x n + 2 v2 = co + e i t + " " + c,,_~(~)
r~--2
. (34)

n-I-ii COS r(Xj -- Xj+I) (24) The coefficients are determined from the boundary
conditions of z, which is calculated from the initial
and final position. Figure 12 shows that tile optimal
tau r(xi - - J2i+l) "4"-t a n ( 1 - r)(Xi_ 1 - - Xi)
fi ~ three point model with 2 trailers being controlled by
di-2 c o s x n + 2
the time polynomial inputs. Also shown in Fig. 12 is
n+l COS r(.Xj -- X j + i ) the result of the S0rdMen's model with the same open
(25)
loop control strategy. 'File both nlodels exhibited a
3:z
similar response to the open loop control.
(i=4,..,,n+l)
tan(1 - r ) ( x . + l - x.+2)
fn+2 = (~6)
d n COS a~n+ 2
f n + 3 ---- t, a n x n + 2 (27)
which yiehl,

1
0 .r "e di cos a.,,+,, cos'-'(x=, - x3)
,~+1 COS r(d'j -- J ' j + l ) SordMen's model T h r e e l)ohlt, model(r = 0.5)
(28)

inl0at tnl~
Of; - r

Oxi- j di-2 c o s a , + 2 cos'-( 1 -- r ) ( x , _ 1 -- x i )

,,+1 cos,'(a'j - "'~+A!
• II (29)
5 t a 2 ~ g

(i=4,....n.+l) S, m d a | e n s model T h r e e point, m o d e l ( r = 0..5)

Figure 12: Open loop Control wMl polynondal input.

252

6 Prototyping and Experiments
6.1 Prototyping
In this subsection we propose some mechanisms for
implementing optimal three-point model. In Example
l -l.ill
~
~ ~
Rodof~
: tt'fdleri+l !
Figure 13: Mechanism of three point model.
1 of Vig. 13, gears A and B are fixed t,o rods i and
i + l, respectively. The body of trailer i is fixed to
rod i. If rod i rotates, rod i + 1 also rotates in the
opposite direction. If the radius of gear A is k times
longer than that of B, we have r = 1/(1 + k). When
radii of gears are equal, r is 0.5. If r is limited to 0.5,
Exampl< 2 is also frasible. Mechanism D is fixed to E
and slides on the axle. Pins F and G are to comlect t o
rods i and i + 1 and slide on E. We adopted Example 2
to avoid mechanical complexity and backrnsh due to
gear trains. Each intermediate trailer can be switched
between r = 0 and 0.5. This allows us to compare path
following properties of the optimM three-point model,
the Serdalen model and the ITS model. Trailer n has
an actuator that rotates rod n about the orogin of
frame n.
6.2 Experiments
We fablieated the prototype of the optimal three
point, model. Experiments of forward and backward
path following were carried out with the prototype.
Figure 14 shows photos of the prototype. The proto-
type consists of a tractor and five trailers. The top
two I)hotos in Fig. 14 show the structure of the first.
through fourth trailers, The bottom two show the
fifth one. The experiments correspond to Example 2
of the previous subsection. Figure 15 shows the opti-
Fib, . . . . . . . . . . ,. . . . ~,
real three-point model following a connected reference
o f linear and circular trajectries. Figure 16 s h o w s tile
backward motion, where trailer n is steeered as men-
tioned in subsection 4.3 to follow a circular reference
trajectory, Note that. the trailer system keeps n o n -
z i g z a g motion.
' ~/l g o d of
" traileri
G ~
Figure 15: Experiment of forward motion.
Figure 16: Experiment of backward m o t i o n .
'7 Conclusion
This paper is summarized as follows:
(1) As a basis to study trailer systems that show both
path-following performance and the chained form con-
vertibility, a kinematic model of trailer systems with
steering is formulated.
(2) We proposed a passive steering mechanism named
the three-point model. Simulation results showed
that the optimal three-point model wMl steering gain
r = 0.5 is superior to known typical industrial trailer
systems in path following perforniance.
(3) Path following stal)ility was compared and dis-
cussed anlong the industrial trailer designs and lhe
optimal three-point model, It is shown by nmnerical
analysis and theoretical development that only the op-
timal t,hree-point model can trace the path of the front
car in both forward and backward directions.
(4) The steering mechanism of the three point model
was proved to make lhe system differentially fiat and
convertible to the chained form.
(5) We designed and fabricated a prototype of the op-
timal three-point model with simple mechanisnl. Ex-
n
periments verified properties that we discussed in this
paper.
253
Acknowledgment A Chained Form Conversion[7]
T h i s research was s u p p o r t e d by t h e " R o b o t Brain
P r o j e c t , " C R E S T P r o g r a m of J a p a n Science and T h e chained form
Technology C o r p o r a t i o n , J a p a n .
:il = v~ (35)
References
d2 = v~ (36)
[1] J. P. Launmnd, "Feasible trajectories for mobile ii = z i - l v l (i = 3 , . . . , n ) (37)
robots with kinematic and environment, constraints",
in Proc. lnternatiotml Conference on Intelligent A uti- is a canonical form of a class of driftless nonholo-
nomous Systems, pp, 346-354, 1986. nomic s y s t e m s , with two i n p u t s and n d i m e n s i o n a l
[2] J. P. Laumond and T. Sim6on, "Motion planning for s t a t e space. All t h e s y s t e m s c o n v e r t i b l e to the chained
a two degrees of freedom mobile robot with towing", form are c o n t r o l l a b l e .
i~t Proc. h~tern.tional Conference on Control and Ap- It is difficult 1.o find a conversion from a. k i n e m a t i c
plicatons, 1989. model to the chained form. We first convert it i n t o a
[3] R. Murra.y and S. Saslry, "'Nonholonomie motion t r i a n g u l a r s t r u c t u r e . Namely,
planning: Sleering using sinusoids', lEEE Tr, ns-
actions on Automatic ('ontrol, vol. 38, pp. 700 716. & = ul (38)
1993. ~d'e = u'e (39)
[4] O. J. Sorda.len. "Teedback control of nonholonomic a;i = f i ( x i - ~ ) u l (i = 3 , . . - , , ) (40)
mobile robots", Ph.D thesis ITK-rapport 1993:5 W,
the Norwegian Inslitule of Technology, 1993. where
def
[5] M. Sampei, "A control strategy for a class of non- x_j. = [ x i , " - , x , ] z (41)
holonomic systems - time-state control form and its
f i ( x i - t ) dS [ f i ( x i - 1 ) , " ' ' , f . . ( x , , - 1 ) ] 7~ (42)
application : , in Proc. 3&'d Conference on Decision
and Control, pp. 1120-1121, 1994.
where f i t ' ) is a s m o o t h function. A s s u m e t h a t at x = p
[6] C. Samson, "Time-varying feedback stabilization of on the c o n f i g u r a t i o n m a n i f o l d we have
nonholonomic car-like mobile robots", Technical Re-
port 1515, INR1A-Sophia Antipolis, September 1991. Ofi(xi-1)
0~_t I,:p # 0 (i = a , . . . , , ) (43)
[7] O. J. S0rdalen and K. Y. WicMund, "Exponential sta-
bilization of car with n t,railers", in Proc. 32th Con-
fevence on Decision and Control, pp. 978-983, 1993. T h e n , at the n e i g h b o r h o o d of x = p, a c o o r d i n a t e
t r a n s f o r m a t i o n from t h e t r i a n g u l a r s t r u c t u r e to the
[8] Y. Nakamura O. J. So rdalen and W. J. Chung, "De-
sign of a nonholonomic manipulator", in Proc. 1994 chained form is given by
l E E S International Conference on Robotics and Au-
tomation, pp. 8--13, 1994, z. = h,(x,,) (44)

[9] W. J. Chung and Y. Nakamura, "Design of the

chained form manipulator", in P~vc. t997 IEEE In-
zi = h.~(x_k)de_fOhi+l(xi+l)
0a'i+l fi+l(a;1 ) (45)

ternational (.;onfi'rencc on Robotics and A,~tomation, def

pp..155-461, 1997. zl = h i ( x , ) = xt (46)
[10] K. Yamamiya el. al., ":I'raih.r", .Iapan Patenl Office (i= 2,...,~-- 1)
Unexamined Patent (~azetle B 6 2 D I 3 0 0 1t2 18928 vl = ~q (47)
(in Japanese). 1990.
Oh.(a~.,) Oh2('r--Z) (4S)
[11] O. Gehring and lf. Frilz, "Laleral control concepll,s for
lruck platooning in the chauffeur project", it~ Proe.
~¢th l'l'orld ('ogress (m ITS, 1997.
where, h,,(x..) is any s m o o t h function which satisfies
[12.] 5'. Umetani and S. llirose, 'q~iomechanical study of
active cord mechanism with tactile sensors", in Proc. 0/,,~ (.,,)
6th lnter~mtional Symposium on Industrial Robotics, a.,, I.,.:. ¢ 0 (49)
pp. c l , l - 10, 1976.
[13] R. Murray D. Tilbury and S. Sastry, "Trajectory gen- In this p a p e r , we choose hn = 3% to o b t a i n the sim-
erMion for the n-trailer problem using goursat nor- plest e q u a t i o n .
real form", in Proc. 32th (:onferencc on Decision and
(.'omrol, 1993.

254