Вы находитесь на странице: 1из 7

Proceedings of the 2006 IEEE/RSJ

International Conference on Intelligent Robots and Systems


October 9 - 15, 2006, Beijing, China

Attitude Control of a Quadruped Trot


While Turning
Luther R. Palmer III

David E. Orin

Department of Electrical and Computer Engineering


The Ohio State University
Columbus, Ohio 43210
Email: palmer.216@osu.edu

Department of Electrical and Computer Engineering


The Ohio State University
Columbus, Ohio 43210
Email: orin.1@osu.edu

Abstract During a complete running stride, which involves


signicant periods of ight during which no legs are contacting
the ground, a quadruped cannot employ static stability techniques.
Instead, the corrective forces necessary to maintain dynamic
stability must be applied during the short stance intervals inherent
to high-speed running. Because of this complexity and the large
coupled forces required to run, much of the research on the control
of quadruped running has focused on planar systems which are not
required to simultaneously control attitude in all three dimensions.
The 3D trot controller presented here overcomes these and other
complexities to control a trot up to 3.75 m/s, approximately
3 body lengths per second, and turning rates up to 20 deg/s.
The biomimetic method of banking into a high-speed turn is also
investigated here. Along with the details of the attitude control
algorithm, a set of control principles for high-speed legged motion
is presented. These principles, such as the need to counteract
the disturbance of swing leg return and the usefulness of force
redistribution during stance, are not dependent on a particular
scale or actuation scheme and can be applied to a wider range of
legged systems.

I. I NTRODUCTION
Legged systems offer many advantages over wheeled vehicles, particularly over broken terrain. A legged system can
use obstacles as footholds in order to climb or jump, rather
than navigate around them as wheeled vehicles must often do.
The potential of legged locomotion is evident in nature as the
cheetah can dynamically maneuver at extremely high speeds
over uneven terrain and as dogs are used to search through
dense wreckage for potential survivors. At the present time,
legged systems are simply not fast enough to be useful for tasks
like military reconnaissance and time-critical search and rescue
operations. To address the issue of speed, this paper presents an
algorithm for high-speed trotting and a set of useful principles
that can be applied to the control of quadruped systems of
different scales and actuation schemes.
In nature, cursorial quadrupeds select gaits at various speeds
based mainly on energy considerations. Although many animals
gallop at top speeds, there is a signicant range of intermediate
speeds for which trotting is the most energy efcient gait [1],
[2]. A number of researchers have studied the quadruped bound
as a high-speed dynamic gait [3], [4], [5], but very few animals
naturally bound. Instead, trotting is often used by animals as
the precursor to galloping. Figure 1 shows the complete trotting

1-4244-0259-X/06/$20.00 2006 IEEE

Fig. 1. Trotting stride showing two stance phases interleaved with two ight
phases. The quadruped is largely uncontrollable during ight [9].

stride with two ight phases, during which the body is largely
uncontrollable, and two stance phases, during which the legs
can apply corrective forces onto the body. Also notice that
diagonal leg pairs are synchronized throughout the entire stride.
During the short intervals of stance, compliant elements
and high-power actuators combine to perform a complex interchange of potential and kinetic energy. The high forces
necessary to overcome frictional and contact losses, and to
redirect the vertical momentum of the body, are strongly coupled to rotations about all three axes in a 3D quadruped. This
coupling is difcult to manage for control purposes because of
the many degrees of freedom. The attitude of the quadruped
becomes largely uncontrollable during ight, and the required
leg movements in preparation for touchdown cause undesired
effects on body motion, especially pitch. The control problem
is further complicated by the fact that the inter-limb and intralimb coordination and actuation schemes that animals use to
run at high speeds are not well understood.
Because of these difculties, much of the previous work on
quadruped control has only considered motion in the sagittal
plane (forward, vertical, and pitch motion) [6], [7]. Algorithms
have also been implemented on quadrupeds limited to move in
steady state trajectories, or that operate around a single speed
[4], [8]. These results show that the highly nonlinear quadruped
system can be linearized and successfully controlled around a
xed point, but these controllers only maintain stability when
the system is well initialized and without the presence of
signicant disturbances. Other assumptions such as massless
legs have also been made to simplify the quadruped system for
control.
This paper presents a control algorithm that overcomes
the above difculties to dynamically stabilize a high-speed
quadruped trot. Dynamic stability here is described as sustained

5743

Authorized licensed use limited to: Khajeh Nasir Toosi University of Technology. Downloaded on December 21, 2009 at 05:40 from IEEE Xplore. Restrictions apply.

tracking of desired inputs and the ability to compensate for


moderate disturbances. A hybrid control strategy divides the
effort into control outputs that are updated once per step, such
as those necessary for velocity and turning rate control, and
control outputs that are continuously updated, such as those
necessary for pitch and roll control. The quadruped system runs
over a range of speeds up to 3.75 m/s and turning rates up to
20 deg/s.
This paper not only presents a successful controller for a
high-speed quadruped trot, but it also introduces several useful
principles of high-speed quadruped motion. These principles
are the more useful result of this work, as they can be applied
to a wider range of legged systems. These principles are as
follows: 1) swing legs can cause signicant disturbances to the
system and their effects should be counteracted, 2) the large
leg forces necessary to reverse the bodys vertical momentum
during stance cause high-frequency pitch and roll motions
that must be continuously controlled during intervals of footground contact, 3) if the body is in a predictable and repeatable
conguration during stance, then velocity, heading, and height
control only need to be updated once per step, and 4) the large
leg forces generated during stance can be redistributed to effect
the desired control in some directions, while maintaining the
natural dynamics in other directions.
This paper is organized as follows. After the quadruped
model is presented in Section II, Section III presents the control
approach for trotting which uses the above principles as its
foundation. The simulation results of the controller are shown
in Section IV, and are followed by a summary and discussion
of the future work.
II. Q UADRUPED M ODEL
A model of the 3D quadruped system used in this work is
shown in Fig. 2. A dynamic model of the leg is shown in
Fig. 3. The legs each have two actuators at the shoulder/hip
joints, one for abduction and adduction of the leg and another to
swing (protract/retract) the leg. The resultant leg angles are a
and s respectively. The abduction and adduction actuators and
axes will hereafter be referred to as ab/ad. Another actuator
in series with a passive spring at the knee can adjust virtual
leg length, r, during ight and continuously adjust the spring
length during stance to add or remove energy from the system.
Energy is stored in the spring as the knee bends during the rst
half of stance and returned to the system as the leg lengthens.
The energy conserved by the spring reduces the effort required
to maintain a consistent ballistic trajectory of the quadruped
during ight. A single linear spring in the leg has been shown
to behave much like an animals complex system of muscles,
tendons and ligaments [10], [11]. The angle of the knee is k ,
and the angle of the virtual leg with respect to the body normal
is l .
Figure 2 shows the specic body angles to be controlled:
roll, , pitch, , and yaw, , along with the forward velocity,
vbx , lateral velocity, vby , and height, h = pzE , at the apex of the
ight phase. The peak height is recognized when the vertical
body velocity is zero and occurs at a time referred to as top of

Fig. 2.

Body model with simplied legs.

Fig. 3.

Dynamic leg model.

ight (TOF). A step starts at a TOF and concludes at the next


TOF, with a diagonal leg pair having contacted the ground and
exerted an appropriate impulse during that time.
The quadruped weighs a total of 68 kg and stands 60 cm high
with the knee springs in their nominal position. The shoulder
separation is 35 cm and the shoulder to hip distance is 1.2 m.
The size was chosen to match those of a mid-sized goat. The
mass of the thigh, mt , and shank, ms , are both 1.0 kg. The
thigh and shank are modeled as slim rods of length 35 cm
with geometrically-centered mass. The spring constant is xed
at 25, 820 N/m.
The control algorithm developed in the next section is
designed for a wide range of body scales and leg congurations
(with and without springs). The algorithm solves for knee
torques, k , swing torques, s , and ab/ad torques, a , which can
be resolved by many different spring/actuator congurations on
each joint.
III. C ONTROL A PPROACH
The leg thrusts delivered onto the body during stance are the
main engines of quadruped motion. The angle of each leg with
respect to the body determines the direction of the impulse
that each leg will produce. This impulse can be directed to
produce corrective lateral and fore-aft forces, as well as angular
moments.
For each step, the main task consists of placing the legs in
a good conguration in preparation for touchdown and then
redistributing the leg forces during stance to deliver desired
moments and forces in some directions without disturbing
the dynamics in other directions. The leg motions during a
stride have been organized into the phases of a discrete state
machine as shown in Fig. 4. During protraction, the swing
angle, s , and ab/ad angle, a , are actuated from their liftoff
positions to desired angles for touchdown. Before the leg can
protract, it must be shortened to avoid stubbing the toe. After
protraction, the leg is lengthened in preparation for touchdown.

5744
Authorized licensed use limited to: Khajeh Nasir Toosi University of Technology. Downloaded on December 21, 2009 at 05:40 from IEEE Xplore. Restrictions apply.

Fig. 5. Attitude control system. Yaw rate, forward velocity, and height are tracked through the step controller which runs once per step while pitch and roll
are regulated through continuous force redistribution during stance.

PROTRACT
SHORTEN

LENGTHEN

Flight
Stance

liftoff
THRUST

Fig. 4.

touchdown
COMPRESS

Leg controller states.

Foot touchdowns for diagonal leg pairs are synchronized by


extending the legs appropriately to compensate for body pitch
and roll.
The spring compresses and stores energy as the knee bends
during the rst phase of stance. When the spring is maximally
compressed, additional energy, E, is added to the spring to
overcome contact losses. Thrust is then delivered onto the body
as spring potential is converted back to vertical velocity until
the foot breaks contact with the ground, and the process repeats.
The step controller, shown in Fig. 5 and described below,
will assign the appropriate leg congurations during ight. A
discussion of the force redistribution during stance will follow.
A. Step Control
The main component of the step controller is a direct adaptive
fuzzy controller. A formal discussion of the fuzzy controller
is not presented here because of space considerations, but a
description of fuzzy control and how it can be applied to
quadruped motion can be found in [12]. The fuzzy controller
receives as its inputs the desired turning rate, d , desired
forward velocity, vdx , desired lateral velocity (always zero), vdy ,
desired height, hd , and the present body state, x. The controller
outputs a set of desired leg touchdown angles, td , and the
amount of energy, E, that will be added to each knee spring at
maximum compression during stance.
Forward velocity is controlled by varying the virtual leg
angle, l , at touchdown. During stance, the fore-aft motion will
be dictated by the natural dynamics of the system. To accelerate
and decelerate, the direction of the total fore-aft impulse can
be chosen by setting appropriate touchdown angles for the
virtual leg. This is similar to Raiberts monopod velocity control
algorithm [6]. The ab/ad angles, a , at touchdown will dictate
the lateral forces and yaw moment and the resulting motion as
shown in Fig. 6. Lateral motion is produced by placing both
legs toward one side of the body. Turning is achieved by placing

(a) Lateral motion.

(b) Yaw motion.

Fig. 6. The ab/ad angles at touchdown cause lateral ground reaction forces
to produce lateral motion as in (a) and yaw motion as in (b).

the front legs to the outside of the turn and the hind legs to the
inside.
The response of the body during stance to the provided
touchdown conguration and energy addition depends on the
tilt angles of the body, namely pitch and roll. The online
learning done by the adaptive fuzzy controller assumes that the
body tilt angles at touchdown and during stance are predictable
and repeatable. During ight, the step controller outputs a
commanded pitch, c , pitch rate, c , roll, c , and roll rate, c ,
for the body during stance. The commanded pitch and pitch rate
are nominally zero, but can be changed to match the inclination
angle on sloped terrain.
Animals have been observed to bank into turns, as shown
in Fig. 7, to keep their ab/ad angles with respect to the body

small. For a given forward velocity, vbx , and turning rate, ,


the legs apply a net force on the body in the transverse plane
at an angle with respect to the earths vertical axis. Conical
pendulum analysis can be used to compute this angle, , by

tan() = (vbx )/g,

(1)

where g is the gravitational acceleration. If leg thrusts along


the leg were the only forces on the body, the ab/ad angles with
respect to the vertical should match . As will be seen in the
next section, ab/ad torques are used during stance to redistribute
the leg forces. The force acting at the conical pendulum angle
is then the resultant force of the leg thrusts along the leg
and the forces normal to the leg generated by ab/ad torques.
Because the ab/ad torques produce forces that are small relative

5745
Authorized licensed use limited to: Khajeh Nasir Toosi University of Technology. Downloaded on December 21, 2009 at 05:40 from IEEE Xplore. Restrictions apply.

(a) Flat turn.


Fig. 7.

(b) Banked turn.

Turning with and without body roll.

0.0 m/s

1.0 m/s

bank angle (deg)

20

20
ab/ad angle
conus angle

10

10

10

10

20

40

20
40 20

3.0 m/s
bank angle (deg)

2.0 m/s

20

10

20
40 20

Biomechanics studies show that pitch stability for the


quadruped trot is achieved by redistributing the vertical impulses during stance between the fore and hind limbs [13].
Previous work has been done on force distribution in legged
robots [14], [15], but this was for walking machines that did
not need the large forces required to run and at times had four
legs in contact with the ground. Another unique aspect of this
work is that only motion on the pitch and roll axes are directly
modied. The forces on the body produced by the large axial
leg thrusts are redistributed while adding small ab/ad torques
to effectively control pitch and roll without causing signicant
disturbance to the forward, vertical, lateral, and yaw motion.
The motion in these four axes, dictated by the outputs of the
fuzzy controller, is not affected by the stance algorithm.
The pitch control element of this controller presents a way
to adjust the axial leg forces to correct pitch error without
changing the systems vertical motion. Because adjusting the
leg forces to correct the pitch motion has an undesired effect on
roll, an algorithm is developed to negate these undesired effects.
The roll element of the algorithm is similar to what has been
described in [16], but is presented more formally here.
1) Pitch Control: The force, f i , that leg i exerts on the body
(ignoring leg inertial forces) is given as

20

10

40

20
40 20

4.0 m/s

20

40

5.0 m/s

20

20

20

10

10

10

10

10

10

20
20
20
40 20
0
20 40 40 20
0
20 40 40 20
0
20 40
yaw rate (deg/s)
yaw rate (deg/s)
yaw rate (deg/s)

Fig. 8. Comparison between ab/ad angles and the conical pendulum angle.
The trained ab/ad angles with respect to the vertical, shown as circles, are very
similar to the computed conus angles, shown as triangles. Both slopes have an
approximate linear relationship with turning rate at a single velocity, but the
slopes diverge from each other as speed increases.

to the leg forces, the ab/ad angles with respect to the vertical
are still very similar to the conus angles, as shown in Fig.
8. Equation 1 was used to calculate the conus angles, shown
as triangles, for 7 turning rates at 6 forward velocities. The
fuzzy controller outputs the desired touchdown ab/ad angles
with respect to the vertical. These are shown as circles. Both
slopes have an approximate linear relationship with turning rate
at a single velocity, but the slopes diverge from each other
as speed increases. From the discrete circled data in Fig. 8,
a continuous equation for ab/ad angles can be developed for
all velocities and turning rates. To keep the ab/ad angles with
respect to the body small, the body roll angle is commanded
to match the ab/ad angles with respect to the vertical.
The next section on force redistribution shows how the
commanded pitch and roll motions are achieved during the
stance phase.

f i = J ( i )T i

where J ( i )T is the transpose of the inverse Jacobian relating


joint torques to forces on the body and i = [a,i , s,i , k,i ]T .
For simplicity, let by J i = J ( i ). The moment, ni , that leg i
exerts on the body is given as

a,i
ni = s,i ,
(3)
c,i
where a,i is the ab/ad actuator torque, s,i is the swing actuator
torque of leg i, and c,i is a constraint torque about the bodys
z b axis. The total moment on the body, nb , is given by
4



ni + i (bpi f i ) .
nb =

(4)

i=1

where i is 1 for legs in contact with the ground and 0 for legs
in ight. The vector b pi gives the position of leg i relative to
the body. With only diagonal leg pair (1,4) on the ground, this
equation becomes
nb = n1 +

B. Force Redistribution
At most two legs simultaneously interact with the ground
during high-speed quadruped trotting, eliminating the possible
use of static stability techniques. Running also requires much
larger leg forces than walking, and these forces are strongly
coupled to all six directions of motion. This coupling motivates
the need for continuous control of the body tilt angles, pitch
and roll, during stance.

(2)

p1 bR1 J T
1 1 + n2


.
+ n3 + n4 + bp4 bR4 J T

4
4

(5)

The matrix b Ri gives the transformation from leg coordinates


to body coordinates. While in stance, the knee torque produces
the dominant pitch on the body. During ight, the swing torque
used to protract the leg delivers an equal and opposite pitch
moment on the body according to Eq. 3. The swing actuators
of the stance legs are not active.

5746
Authorized licensed use limited to: Khajeh Nasir Toosi University of Technology. Downloaded on December 21, 2009 at 05:40 from IEEE Xplore. Restrictions apply.

The pitch component of the previous equation can now be


written as:
nyb = y1 k,1 s,2 s,3 + y4 k,4

where i is a vector that relates the knee torque of leg i, k,i ,


to body forces. The variables to adjust knee torque, k,1 and
k,4 , are added and the equation becomes

(6)

where i is a vector that relates the knee torque of leg i, k,i , to


body moments. Notice that control principle 1 will be adhered
to by retaining the swing torques of the ight legs in this
equation for body pitch moment. Results for pitch stability will
be shown with and without this swing torque compensation.
Adding variables to adjust the knee torques by k,1 and k,4 ,
the equation becomes

0
0
fbz = z1 (k,1
+ k,1 ) + z4 (k,4
+ k,4 ) mg .

Again, the new force, Efbz , is the sum of the original force,
plus the effects of the change in knee torques:

E z,0
fb ,

fbz =

y
y
nby = ny,0
b + 1 k,1 + 4 k,4 .

nby

,
= kp (c ) + kd ( c )

(9)

an appropriate k,1 and k,4 can be solved for by setting


Eq. 8 equal to Eq. 9, which produces:
y,0

y1 k,1 +y4 k,4 = kp (c )+kd ( c )n


b . (10)

With this equation, a k,1 and k,4 are produced to deliver


a corrective pitch moment on the body, but these changes in
knee torques should not disturb the systems vertical dynamics.
Thus, another equation for k,1 and k,4 must be derived
that satises this constraint.
The total force on the body in earth coordinates, Ef b , is given
by
fb =

4


i ERi J T
i i + mg .

(11)

i=1

where g = [0 0 g]T and ERi is the transform from leg i to


earth coordinates. With diagonal leg pair (1,4) on the ground,
this equation can be written as
T
E
f b = ER1 J T
1 1 R4 J 4 4 + mg .

E z
fb

= z1 k,1 + z4 k,4 mg ,

(13)

(15)

E z,0
fb

(16)

and k,1 and k,4 are constrained by


z1 k,1 + z4 k,4 = 0 .

(17)

Solving Eq. 10 and Eq. 17 simultaneously to nd k,1 and


k,4 produces:

y1
z1

y4
z4


k,1
k,4


ny,0
kp (c ) + kd ( c )
b
=
. (18)
0

The k,1 and k,4 produced by this set of equations effectively change the pitch moment of the body without affecting
the vertical dynamics. When there is no pitch or pitch rate
error, the stance legs simply negate the moment caused by the
swing legs. As the quadruped trots faster, swing leg torques
must increase to generate faster protraction trajectories. Without
compensating for the disturbances on the body caused by swing
torques, pitch will become unstable at high speeds.
2) Roll Control: As mentioned before, modied knee
torques produce an unwanted change in the roll dynamics.
A controller must be developed for the roll axis that negates
the effect of this unwanted moment. The roll moment to be
negated is computed, added to a stance roll controller, and then
delivered to the system by applying torques to the ab/ad axes
of the two stance legs. This is the rst constraint on the ab/ad
torques, a,1 and a,4 . These torques (plus the redistributed
knee torques) should not produce an unwanted yaw moment,
so a second constraint equation must be derived such that the
natural yaw dynamics are not affected by either the pitch or
roll controller.
With diagonal leg pair (1,4) on the ground, the roll moment
on the body is given by:
nxb = a,1 + x1 k,1 a,2 a,3 a,4 + x4 k,4 ,

(12)

Again, knee torques produce the dominant forces on the body


in the earths z E direction, so Eq. 12 can be written as

+ z1 k,1 + z4 k,4 .

fbz =

(8)

Constraining the new pitch moment to be equal to a commanded moment from a linear controller,

E z,0
fb

Because we desire no change in the vertical dynamics,

0
0
nby = y1 (k,1
+ k,1 ) s,2 s,3 + y4 (k,4
+ k,4 ) , (7)
0
where k,i
is the torque produced by the passive spring at
the knee without movement of the spring anchor. This torque
naturally increases during leg compression as the spring absorbs
energy and decreases back to zero as the spring returns
all of

y
its energy to the system. The new pitch moment, nb is given
as the sum of the original pitch moment ny,0
plus the pitch
b
moment produced by k,1 and k,4 :

(14)

(19)

where xi is the coefcient that relates the knee torque of


leg i to the moment about the roll axis. Because the roll
controller works after the pitch controller, the knee torques in
this equation, k,1 and k,4 , are equal to the new knee torques,

5747
Authorized licensed use limited to: Khajeh Nasir Toosi University of Technology. Downloaded on December 21, 2009 at 05:40 from IEEE Xplore. Restrictions apply.

0
0
k,1
+ k,1 and k,4
+ k,4 , respectively. Again, ight leg
torques, a,2 and a,3 , that affect body motion are accounted
for. Setting the roll moment of Eq. 19 equal to a desired roll
moment, from a linear controller, results in the rst constraint
equation for a,1 and a,4 :

a,1 + a,4 = x1 k,1 a,2 a,3 + x4 k,4


kp (c ) kd ( c )
. (20)

Fig. 9.

The body moment on the yaw axis is given by


=

z1 a,1

z1 k,1

z4 a,4

z4 k,4

velocity (m/s)

nzb

(21)

where i is a vector that relates the ab/ad torque of leg i,


a,i , to body moments. Because the additional knee torques,
k,1 and k,4 , together with the ab/ad torques should create
no moment on the yaw axis, our second constraint equation is
given by
+

z4 a,4

z1 k,1

z4 k,4

desired
actual

3
2
1
0
1
0

(22)

Solving Eq. 20 and Eq. 22 simultaneously produces


These equations produce ab/ad torques, a,1 and a,4 , such


that the roll and yaw moment caused by the pitch controller
is negated, and the roll moment caused by the swing legs is
negated. Note that similar equations for pitch and roll control
can be developed when legs 2 and 3 are on the ground.
Through the pitch and roll controllers, the leg forces during stance are redistributed to correct pitch and roll errors
without affecting the natural vertical and yaw motions that
will result from the fuzzy control. Because pitch and roll are
well regulated, the quadruped will be in a predictable and
repeatable conguration at touchdown and during stance, which
is necessary for velocity, heading, and height control.
IV. R ESULTS
The control algorithm was tested in a fully-dynamic simulation environment [17]. System losses are modeled as damping
in the compliant ground. Ground spring and damping coefcients are 75 kN/m and 2 kN/m respectively. Ground static
and kinetic friction coefcients are 0.75 and 0.6 respectively,
matching the properties of rubber on concrete. The system in
simulation is shown in Fig. 9.
Figure 10 shows the response of the quadruped to changes
in desired speed and turning rate. Acceleration is limited to
0.25 m/s per step, allowing the quadruped to increase velocity
by 1 m/s in four steps. The turning rate can increase or
decrease by 10 deg/s in one step. The top speed achieved
using this controller is 3.75 m/s. At speeds higher than this, the

40

60

80

100

desired
actual

20
15
10
5
0
5
0



1
1
a,1
=
z1 z4
a,4

x
1 k,1 a,2 a,3 + x4 k,4

kp (c ) kd ( c )
. (23)
z
z
1 k,1 4 k,4

20

25
yaw rate (deg/s)

z1 a,1

Quadruped in full 3D simulator.

Fig. 10.

20

40

60
step

80

100

Tracking of velocity and turning rate.

sweep angle for protraction becomes very large and the moment
on the body created by the resultant swing torques cannot be
overcome. At very low speeds, swing leg compensation is not
needed to maintain pitch stability. As speed increases and more
swing torque is needed to return the leg during protraction, the
algorithm must compensate for the undesired effect.
Figure 11 shows the pitch motion of the quadruped running at
3.0 m/s with and without swing torque compensation. During
ight, the body is uncontrolled and body pitch error increases
based upon the pitch rate at liftoff and the dynamic swing
effects of all four legs. During stance, the pitch is controlled
back toward zero. Without compensating for the swing torques,
the pitch motion during stance reaches an angle of zero but has
nonzero rate. This steadily causes the pitch angle to diverge
toward instability. With compensation, the pitch excursions
during ight remain bounded because the pitch rate at liftoff
is near zero. This results in predictable and repeatable tilt
angles during stance, which improve the performance of the
step controller. The maximum speed achieved without swing
torque compensation is 2.75 m/s for straight ahead locomotion.
This speed is limited further when a turning rate is desired.
The ability of the quadruped to turn at high speeds is directly
related to the success of the stance controller to keep the tilt
errors small.
V. S UMMARY AND F UTURE W ORK
A hybrid control system using a discrete controller running
once per step and a continuous controller during stance was
presented that dynamically stabilized a quadruped running at
3.75 m/s and turning at 20 deg/s. These speeds and turning

5748
Authorized licensed use limited to: Khajeh Nasir Toosi University of Technology. Downloaded on December 21, 2009 at 05:40 from IEEE Xplore. Restrictions apply.

No Swing Leg Compensation

and the correlation between the ab/ad leg angles and the conical
pendulum angles. The present system and controller can also
serve as a testbed for use of other biomimetic strategies such
as early leg retraction and the use of neuro-oscillators. We also
expect that elements of this controller will be useful for other
high-speed quadruped gaits such as the canter and gallop.

pitch (deg)

4
2
0
2
4
0

pitch
stance
0.5

1.5

2.5

3.5

With Swing Leg Compensation

pitch (deg)

ACKNOWLEDGMENTS

Support was provided by grant no. IIS-0208664 from the


National Science Foundation to The Ohio State University, and
through a teaching associateship in the Department of Electrical
and Computer Engineering at The Ohio State University.

0
2
4
0

pitch
stance
0.5

1.5

2
time (s)

2.5

3.5

Fig. 11. Running at 3.0 m/s with and without swing leg compensation.
Without swing leg compensation (top gure), body pitch oscillates between
large angles and quickly goes unstable. With swing leg compensation (bottom
gure), pitch motion is under control. In both gures, legs are in contact
with the ground when the stance signal is high. During ight, the body
is uncontrolled and body pitch error increases. During stance, the pitch is
controlled back to zero.

Fig. 12.

KOLT vehicle at Stanford University [18].

rates were attainable because the force redistribution algorithm


implemented during stance corrects the pitch and roll motion
without impacting the forward, lateral, vertical, or yaw motion.
This provided a predictable and repeatable body conguration
during stance, which improves the performance of the step
controller.
The effect of negating the dynamic effects of swing leg
torques was shown, as well as preliminary ideas on the correlation between conical pendulum analysis and a commanded
body roll angle during high speed turning. More generally, the
paper presented four control principles that can be applied to
high-speed quadruped motion on systems of different congurations and actuation schemes.
The control algorithm presented here is designed for the
Kinetically Ordered Locomotion Test (KOLT) Vehicle [18],
Fig. 12, at Stanford University. This vehicle will face many
challenges not realized by the system in simulation such as
limited processing and actuation power, imperfect sensing,
and unmodeled losses. It is hoped that the robustness of this
algorithm in simulation will lead to effective control given these
practical limitations.
Future work will also involve more detailed study of turning

R EFERENCES
[1] D. F. Hoyt and C. R. Taylor, Gait and the energetics of locomotion in
horses, Nature, vol. 292, pp. 239240, 1981.
[2] K. J. Waldron and P. Nanua, Energy comparison between trot, bound,
and gallop using a simple model, Journal of Biomechanical Engineering,
vol. 117, no. 4, pp. 466473, 1995.
[3] M. Buehler, R. Battaglia, A. Cocosco, G. Hawker, J. Sarkis, and K. Yamazaki, SCOUT: A simple quadruped that walks, climbs, and runs,
in Proceedings of the IEEE International Conference on Robotics and
Automation, (Leuven, Belgium), pp. 17071712, 1998.
[4] M. D. Berkemeier, Modeling the dynamics of quadrupedal running,
International Journal of Robotics Research, vol. 17, pp. 971985, September 1998.
[5] A. Neishtadt and Z. Li, Stability proof of Raiberts four-legged hopper in
bounding gait, Tech. Rep. 578, New York University, September 1991.
[6] M. H. Raibert, Trotting, pacing, and bounding by a quadruped robot,
Journal of Biomechanics, vol. 23, suppl. 1, pp. 7998, 1990.
[7] H. M. Herr and T. A. McMahon, A trotting horse model, International
Journal of Robotics Research, vol. 19, pp. 566581, June 2000.
[8] D. P. Krasny and D. E. Orin, A 3D galloping quadruped robot, in
8th International Conference on Climbing and Walking Robots (CLAWAR
2005), (London, U.K.), pp. 467474, September 2005.
[9] P. P. Gambarian, How Mammals Run. New York: John Wiley & Sons,
1974.
[10] R. Blickhan, The spring-mass model for running and hopping, Journal
of Biomechanics, vol. 22, pp. 12171227, 1989.
[11] C. Farley, J. Glasheen, and T. McMahon, Running springs: speed and
animal size, Journal of Experimental Biology, vol. 185, pp. 7186, 1993.
[12] D. W. Marhefka, D. E. Orin, J. P. Schmiedeler, and K. J. Waldron,
Intelligent control of quadruped gallops, IEEE/ASME Transactions on
Mechatronics, vol. 8, pp. 446456, December 2003.
[13] D. V. Lee, J. E. A. Bertram, and R. J. Todhunter, Acceleration and
balance in trotting dogs, Journal of Experimental Biology, vol. 202,
pp. 35653573, 1999.
[14] F. T. Cheng and D. E. Orin, Efcient formulation of the force distribution equations for simple closed-chain robotic mechanisms, IEEE
Transactions on Systems, Man, and Cybernetics, vol. 21, pp. 2532,
January/February 1991.
[15] K. Yoneda, H. Iiyama, and S. Hirose, Sky-hook suspension control of
a quadruped walking vehicle, in Proceedings of the IEEE International
Conference on Robotics and Automation, (San Diego, CA), pp. 9991004,
1994.
[16] L. R. Palmer III and D. E. Orin, 3D control of a high-speed quadruped
trot, Industrial Robot, vol. 33, no. 4, pp. 298302, 2006.
[17] S. McMillan, D. E. Orin, and R. B. McGhee, Dynamechs: An object
oriented software package for efcient dynamic simulation of underwater
robotic vehicles, in Underwater Robotic Vehicles: Design and Control,
pp. 7398, Albuquerque, NM: TSI Press, 1995.
[18] J. G. Nichol, S. P. Singh, K. J. Waldron, L. R. Palmer III, and D. E.
Orin, System design of a quadrupedal galloping machine, International
Journal of Robotics Research, vol. 23, no. 10-11, pp. 10131027, 2004.

5749
Authorized licensed use limited to: Khajeh Nasir Toosi University of Technology. Downloaded on December 21, 2009 at 05:40 from IEEE Xplore. Restrictions apply.

Вам также может понравиться