Robust Motion Control of a Four Wheel Drive Skid-Steered Mobile Robot

Sercan Arslan
1
and Hakan Temelta
2


1,2
Istanbul Technical University, Control Engineering Department, 34469 Maslak, Istanbul, Turkey
[email protected], [email protected]


Abstract

In this paper robust motion control of a four wheel drive
skid-steered mobile robot (4WD SSMR) is presented. We
have developed a motion control system where a kinematic
trajectory tracking controller based on the vector field
orientation (VFO) strategy and a robust dynamic velocity
controller based on the sliding mode control (SMC)
technique and the computed torque method (CTM) are
combined. Asymptotic stability for a class of reference
trajectories is guaranteed by the VFO method while the
stability of the velocity controller is based on the Lyapunov
theory. In addition to the original VFO method a multi
parameter orienting control is used. A 4WD SSMR is
designed in a three dimensional (3D) realistic simulation
environment to test the performance of the motion control
system developed. Simulation results have shown the
stability and robustness of the motion control system even
under heavy perturbed conditions and the proposed multi
parameter orienting control strategy has the advantage of
smoother path tracking.

1. Introduction

Skid-steered mobile robots (SSMR) are well-known for their
robust structure which is suitable for outdoor usage especially
on rough terrains. SSMRs are differentially driven vehicles
(DDVs) because they are rotated by differential speeds or
torques on left and right side wheels. SSMRs do not have a
mechanical steering system and lateral skid is necessary for the
vehicle to change its heading direction. Because of this nature of
skid-steered vehicles, wheel-ground interaction forces play an
important role in the vehicle dynamics. The most important
ones, lateral friction forces due to sliding of wheels on ground
may be too high on a hard terrain such as an asphalt road. As a
result it may become too difficult to control the yaw rate of the
vehicle and require high torques generated by the vehicle
actuators. Hence power loss due to high friction forces is
inevitable.
Motion control or trajectory tracking for WMRs is often
studied considering only kinematics omitting the dynamic
properties of the vehicle assuming that wheels track commanded
velocities perfectly and they do not slip while rolling. As a
result it is assumed that the nonholonomic constraints of the
vehicle are satisfied all the time during robot motion such that
one can calculate wheel angular velocities for a desired robot
motion. In fact parameter and non-parameter perturbations,
external disturbances, modeling errors due to simplifying
assumptions and unmodeled system dynamics have impacts on
the robot motion directly or indirectly disturbing the slip-skid
phenomena. Hence the wheels to body motion kinematic model
does not hold all the time causing tracking errors. So in this
study we consider a dynamic model based control strategy to
robustly stabilize a class of reference trajectories which rejects
aforementioned effects.
In 1996, Fierro and Lewis introduced a combined kinematic
and torque control framework using backstepping technique to
join robot kinematics into dynamics allowing one to apply
control approaches from model dependent computed torque
method (CTM) to robust sliding mode control (SMC) technique
[1]. In 1999, Caracciolo et al. studied the dynamics of a 4WD
SSMR offering a nonholonomic operational constraint on the
instantaneous center of rotation (ICR) of the vehicle [2]. This
additional nonholonomic operational constraint acts like an
outer-loop controller term preventing excessive skidding of the
vehicle by limiting the vehicles lateral velocity with the yaw
rate [3, 4]. In 2004, Kozlowski et al. redefined the kinematic
and dynamic model of 4WD SSMR in [4] and [5] based on the
model given in [2]. Later Michaek and Kozowski introduced a
novel VFO feedback control method for trajectory tracking of a
DDV in [6]. Also Michaek et al. extended the VFO control
method for the case of limited skid-slip phenomena in [7]. In
2008, a decoupling design approach using two new torque
variables for controlling the linear velocity and yaw angle of a
4WD SSMR platform using SMC technique without the so-
called operational constraint is introduced [8].
Here we propose a robust motion control system for the class
of 4WD SSMR platforms based on the backstepping kinematics
into dynamics framework and VFO trajectory tracking control
method while controlling the vehicles linear and angular
velocities using a CTM plus SMC technique considering both
system dynamics and kinematics.
A 4WD SSMR platform is being developed at the robotics
laboratory in our department. The state of the art computer
aided design (CAD) model of the mobile robot platform is
shown in Fig. 1.
This paper is organised as follows. First the mathematical
model of the vehicle is derived and next, a model based robust
motion control system is designed. Then simulation results are
given. Finally simulation results are analyzed and final remarks
are made.


Fig. 1. 4WD SSMR platform

2. Mathematical Model

The vehicle model is derived based on the assumptions given
in [2] and [4] except that the kinetic energy of wheels is not
neglected here. We consider only planar motion. Each wheel-
ground contact is a single point and normal forces acting on the
wheel-ground contact points are constant depending on the mass
of vehicle and gravity. We assume that wheels do not slip while
rolling. Wheel-ground interaction forces are represented with a
conventional coulomb friction model. Lateral friction is due to
sliding and longitudinal friction is due to rolling of wheels on
the ground. The vehicle body is represented with a point mass
located at the center, near the front side of vehicle. The wheel
torques are distributed equally on each side and servo drives
track torque commands perfectly such that the electric drives
dynamics can be neglected.

2.1 Kinematics

Kinematics of mobile robot platform is illustrated in Fig. 2.
The vehicle configuration vector in global coordinate frames is

| |
3 T
q X Y u = e (1)

where X, Y and are the position and orientation of vehicle
respectively. The transformation between the local velocities
defined in the local coordinate frames attached on the vehicles
center of mass (COM) and the generalized velocities is


cos sin 0
sin cos 0
0 0 1
x
y
X v
Y v
w
u u
u u
u

=
( ( (
( ( (
( ( (
( ( (

(2)

COM is d e (0, b) distance away from the center of
geometry (COG). ICR is located on the axis that intersects COG
as shown in Fig. 2. In this case the nonholonomic operational
constraint that limits lateral skid is defined as [2]

0
y
v dw = (3)
| || | ( ) sin cos 0
T
d X Y A q q u u u = =

(4)

x
y
v
vfl
vrl
vfr
vrr
vflx
vfly
vfrx
vfry
vrrx
vrry
vrlx
vrly
v
x
vy
w
ICR
X
Y
c
c
a
b
COM
d
COG

Fig. 2. 4WD SSMR kinematics


Fig. 3. 4WD SSMR dynamics

Then we can rewrite the Eq. (2) in the form below where
S(q) e
3x2
is a matrix and q e
2
is called as the control input
vector at kinematic level defined as

( ) q S q q = (5)
( )
cos sin
sin cos
0 1
d
S q d
u u
u u

=
(
(
(
(

, | |
T
x
v w q = (6)

and since the columns of S(q) are always in the null space of
A(q) the following expression is satisfied [2, 4].

0
T T
S A = (7)

2.2. Dynamics

Dynamics of mobile robot platform is illustrated in Fig. 3.
First we introduce the left and right side forces then we describe
the wheel dynamics

2
L fl rl
F F F = = , 2
R fr rr
F F F = = (8)

w w
I w DF t = (9)

where I
w
is the wheel inertia, w
w
= [w
fl
w
rl
w
fr
w
rr
]
T
e
4
is the
wheel angular speeds vector, t = [t
fl
t
rl
t
fr
t
rr
]
T
e
4
is the
wheel torques vector, F = [F
L
F
R
]
T
e
2
is the force vector and
D e
4x2
is the force-torque conversion matrix defined as


1 1 0 0
0 0 1 1 2
T
r
D =
(
(

(10)

where r is the wheel radius. Next the equations of robot motion
in global coordinates follow

( ) ( ) cos cos sin
L R x y
mX F F f f u u u = +

(11)
( ) ( ) sin sin cos
L R x y
mY F F f f u u u = + +

(12)
( )
L R r
I c F F M u = +

(13)

where m is the vehicles mass and I is the vehicles inertia about
z-axis, f
x
and f
y
are the rolling and sliding friction forces
respectively and M
r
is the resistive moment about z-axis.
Rolling friction forces are too small when compared to sliding
friction forces in ideal case. A realistic model for rolling friction
in case of slip can be found, for example in [3]. We define the
friction forces for only front-left wheel as follows

( ) sgn
flx x fl flx
f N v = , ( ) sgn
fly y fl fly
f N v = (14)

where normal forces acting on the wheel-ground contact points
due to gravity are calculated as


2
fl fr
b mg
N N
a b
= =
+
,
2
rl rr
a mg
N N
a b
= =
+
(15)

and finally the resistive moment is calculated as follows.


| | | |
| |
r fly fry rly rry
frx rrx flx rlx
M a f f b f f
c f f f f
= + +
+ +
(16)

The general form of the vehicle dynamics including the
nonholonomic constraint using Euler-Lagrange principle and
introducing an additional vector for representing disturbances is

( ) ( ) ( )
T
d
M q q R q F B q F A + + = + (17)

where M e
3x3
is the mass and inertia matrix, R e
3
is the
vector of resistive forces and torques, F
d
e
3
is the vector of
disturbances, B e
3x2
is the input matrix, F is called as the
control input at dynamic level previously defined, A is the
constraint vector as in Eq. (4), and is the vector of Lagrange
multipliers.

( )
0 0
0 0
0 0
m
M q m
I
=
(
(
(
(

, ( )
cos cos
sin sin B q
c c
u u
u u =

(
(
(
(

,
( )
cos sin
sin cos
x y
x y
r
f f
R q f f
M
u u
u u

= +
(
(
(
(

(18)

Taking the time derivative of Eq. (5) yields

( ) ( ) q S q S q q q = +

(19)

and using the relationships Eq. (5), Eq. (7) and Eq. (19) one can
convert the dynamic system in Eq. (17) to


d
M C R F BF q q + + + = (20)

T
M S MS = ,
T
C S MS =

,
T
R S R = ,
T
d d
F S F = ,

T
B S B = (21)
3. Motion Control

Motion control of a vehicle is often regarded as trajectory
tracking control. Here we regard it as the motion control system
that has two sub-systems, that is, trajectory tracking controller at
kinematic level and velocity controller at dynamic level. Such a
control system is illustrated in Fig. 4.

3.1. Control Problem

The control objective is to asymptotically stabilize the
trajectory tracking error at origin. Here we restrict the input
trajectories to be persistently exciting and admissible. Let the
reference trajectory be denoted by q
d
then the trajectory tracking
error is e(t) q
d
(t) - q(t). Now define the admissible local
velocity vector
d
(t) [v
d
(t) w
d
(t)]
T
then the so-called
persistently exciting reference trajectory is defined as

( ) ( ) ( ) ( )
d d d
q t S q t t q , ( ) 0
d
v t > (22)

The trajectory tracking control problem is to find a smooth
velocity control [1]

( ) ( ) , ,
c d
t f e K q q = (23)

such that e(t) 0 as t , where K is the controller design
parameters vector.

3.2. Trajectory Tracking

A velocity control that achieves tracking for a DDV is given
in [1]


1
2 3
cos
sin
xd x
c
d xd y xd
v e k e
w k v e k v e
u
u
q
+
=
+ +
(
(

(24)

where k
1
, k
2
, k
3
> 0 are controller design parameters and e
x
, e
y
,
e
u
are the components of the error vector in local coordinates.
Another approach, VFO is a motivating control technique to
calculate such velocity commands for DDVs. The reader can
refer to [6] (an application of VFO to a DDV) and [7] (an
extension of VFO in the case of skid-slip phenomena) for more
about VFO. In VFO strategy the trajectory tracking control
problem is divided into two subtasks, convergence of position
and orientation to their desired values. The vehicle is driven by
the pushing control v
c
with the careful pushing strategy while
the orienting control w
c
is responsible for matching the
vehicles heading vector with the position convergence vector.
Such tasks are accomplished with the choice of proper
convergence vector field which defines the instantaneous
convergence direction and orientation for the vehicle.



Fig. 4. Motion control system

For control purposes with VFO, we define a new position
vector q
c
[X
c
Y
c
]
T

2
, that is COG, then the new position
tracking error becomes e
c
q
cd
- q
c
where


cos
sin
c
c
X X d
Y Y d
u
u

( (
( (

,
cd c
c
cd c
X X
e
Y Y

(
(

(25)

Definition of the so-called convergence vector field is h
c

[h
c
*
h
o
]
T

3
where h
c
*

2
defines the convergence direction
and orientation of q
c
sub-state and h
o
defines the convergence
orientation of variable

( )
* *
1 c c c d d
h k e v S u = + ,
1 o a a
h k e u = +

(26)

where S
1
*
(u
d
)
2
defines the instantaneous heading direction
of the reference vehicle, e
a
is the auxiliary angle error and u
a
is
the so-called auxiliary angle


( ) | |
*
1
cos sin
T
d d d
S u u u = ,
* T
a c
e T h Z (27)
( )
cos sin
sin cos
T
u u
u
u u

=
(
(

,
*
a c a
h e u u Z + = (28)

and we define the pushing and orienting control commands.


*
cos
c c a
v h e ,
c o
w h (29)

A geometrical VFO tracking scheme for k
c
=k
1
=1 is
illustrated in Fig. 5. For system in Eq. (5) to track a reference
trajectory in Eq. (22) it is sufficient to drive the vehicle with the
velocity control
c
(t) [v
c
(t) w
c
(t)]
T
. This is guaranteed by the
VFO method. The orienting control in Eq. (26) is similar to a
PD plus control but the derivative parameter is fixed to one.
Hence adjusting the controller performance is limited to a single
proportional parameter. Here we propose a PID plus multi
parameter orienting control defined as below.


1 2 3 c a a a
w k e k e k e dt u = + + + }

(30)

COG
y

Xc
Yc
COG
y
x
Xcd
Ycd
Actual Posture
Reference Posture
d
ea
S1
*
()
hc
*
Y
X
ec
ho
wd
vcS1
*
()
x
wc
vdS1
*
(d)

Fig. 5. VFO trajectory tracking scheme


Fig. 6. Velocity control with SMC technique

3.3. Velocity Control

Now we need to construct a robust velocity control system
that will track the desired velocities to complete the motion
control system in Fig. 4. The block diagram of velocity control
system is illustrated in Fig. 6. Applying the following nonlinear
feedback to the system in Eq. (20) we have

| |
1
u M BF C R q

= (31)
( , , ) u n q q q q = (32)
| |
1
w w
DB Mu C R I w t q

= + + + (33)

where n [n
1
n
2
]
T

2
is the vector of uncertainties and the
control law is


d
u K e
q q q
q o = + + (34)
| |
2 T
v w
e e e
q
e , | |
2 T
v w q
o o o e (35)

where K
q

2x2
is a design matrix, e
q
= q
d
- q is the velocity
error and o
q
is the sliding mode control law. Then one can write
the velocity error dynamics

( ) e K e n
q q q q
o = + (36)

The Lyapunov candidate V = (e
v
2
+e
w
2
)/2 is zero only for e
q
=
0 and V 0 for all e
q

2
. Taking the time derivative of V and
using the Eq. (36) yields

| |
T T
V e K e e n
q q q q q
o = +

(37)

and we define the sliding surface s
q
[s
v
s
w
]
T

2
and the
sliding mode control law as

| |
T
v w
s V e e e
q q
= c c = (38)
( ) sgn
v v v
s o = , ( ) sgn
w w w
s o = (39)

1 v
n > ,
2 w
n > (40)

where
v
and
w
are the linear and angular velocity sliding mode
control gains respectively chosen high enough to reject the
disturbances and uncertainties in the system. It is clear that V is
zero only for e
q
= 0 and V 0 for all e
q

2
so the asymptotic
convergence of e
q
to zero is guaranteed.

4. Simulation

A 4WD SSMR is modeled in a 3D realistic environment
using MD-ADAMS View, Tire and Road modules. The
controller is implemented with MATLAB-Simulink interface. A
circular reference trajectory is applied to test various control
systems. The reference vehicle starts at location (2.5m, 0m) with
yaw angle u
d
(0)=t/2rad and moves along a circle with constant
speeds. We assigned the vehicle parameters as m=150kg,
I=30kgm
2
,
x
=0.003,
y
=0.6, r=0.2m, a=0.2m, b=0.3m,
c=0.45m, d=0.05m, the trajectory tracking controller parameters
k
c
=1, k
1
=5, k
2
=1, k
3
=1 and the velocity controller gain K
q
=[10 0;
0 10]. We assigned the sliding mode gains with a formula,

w
=
v
cm/I,
v
=100,
w
=225. In order to reduce the chattering
effects we used low-pass filters, rate limiter and saturation
blocks for F, o
v
and o
w
. Figure 7 shows the results for VFO and
CTM while Fig. 8 shows the results for VFO and SMC based
motion control system. In Fig. 8a trajectory tracking, in Fig. 8b
the desired, commanded and actual velocities and in Fig. 8c the
left and right side control forces are shown.

4 2 0 2 4
2
1
0
1
2
3
4
5
6
X [m]
Y
[m
]


Reference Trajectory
Actual Trajectory
0 5 10 15 20 25 30
1
0.5
0
0.5
1
1.5
2
2.5
t [s]


ex [m]
ey [m]
et [rad]


Fig. 7. VFO and CTM with 30% parameter perturbation

4 3 2 1 0 1 2 3 4
1
0
1
2
3
4
5
6
X [m]
Y
[m
]


Reference Trajectory
Actual Trajectory
0 5 10 15 20 25 30
1
0.5
0
0.5
1
1.5
2
2.5
t [s]


ex [m]
ey [m]
et [rad]

(a)
0 5 10 15 20 25 30
0
0.2
0.4
0.6
0.8
1
1.2
1.4
t [s]


vd [m/s]
vc [m/s]
v [m/s]
v filtered
0 5 10 15 20 25 30
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
t [s]


wd [rad/s]
wc [rad/s]
w [rad/s]
w filtered

(b)
0 5 10 15 20 25 30
200
150
100
50
0
50
100
150
200
t [s]
F
L
[N
]


Actual
Filtered
0 5 10 15 20 25 30
200
150
100
50
0
50
100
150
200
t [s]
F
R
[N
]


Actual
Filtered

(c)

Fig. 8. VFO and SMC with 87.5% parameter perturbation (a)
Trajectory tracking (b) Linear and angular velocities (c) Left
and right side forces
5. Conclusions

In this paper we have developed a robust motion control
system which consists of two sub-systems. In kinematic level
VFO trajectory tracking method with orienting and pushing
strategy is used. In dynamic level robustness is achieved by
using CTM plus SMC technique which fully rejects
disturbances, modeled and unmodeled system dynamics. A PID
plus multi parameter orienting control is used instead of single
parameter dependent one in the original VFO method. A 4WD
SSMR is designed in CATIAv5, Siemens NX6 and simulated in
MD-ADAMS multi body dynamics engine. Motion control
systems are implemented in Simulink and interfaced with MD-
ADAMS. Simulation results proved the stability and robustness
of the motion control system for persistently exciting admissible
reference trajectories even under heavy parameter uncertainty
conditions. We also achieved a smoother path tracking by using
the multi parameter orienting control.

Acknowledgement

This study is supported by The Scientific and Technological
Research Council of Turkey under grant number 110E194.

6. References

[1] R. Fierro, F. L. Lewis, Control of a Nonholonomic Mobile
Robot: Backstepping Kinematics into Dynamics, Journal
of Robotic Systems, vol. 14, no. 3, pp. 149-163, 1997.
[2] L. Caracciolo, A. De Luca, S. Iannitti, Trajectory
Tracking Control of a Four-Wheel Differentially Driven
Mobile Robot, IEEE Int. Conf. on Robotics and
Automation, Detroit, Michigan, pp. 2632-2638, May 1999.
[3] J. Yi, D. Song, J. Zhang, Z. Goodwin, Adaptive Trajectory
Control of Skid-Steered Mobile Robots, IEEE Int. Conf.
on Robotics and Automation, Rome, Italy, pp. 2605-2610,
April 2007.
[4] K. Kozlowski, D. Pazderski, I. Rudas, J. Tar, Modeling
and control of a 4-wheel skid-steering mobile robot: From
theory to practice, Budapest Polytechnic Jubilee
Conference, Science in Engineering, Economics and
Education, Budapest, Hungary, September 2004.
[5] D. Pazderski, K. Kozlowski, W. E. Dixon, Tracking and
Regulation Control of a Skid Steering Vehicle, American
Nuclear Society (ANS) Tenth International Topical Meeting
on Robotics and Remote Systems, Gainesville, Florida, pp.
369-376, March 28 - April 1, 2004.
[6] M. Michaek, K. Kozlowski, Vector-Field-Orientation
Feedback Control Method for a Differentially Driven
Vehicle, IEEE Transactions on Control Systems
Technology, vol. 18, no. 1, pp. 45-65, January 2010.
[7] M. Michaek, P. Dutkiewicz, M. Kieczewski, D. Pazderski,
Vector-Field-Orientation Tracking Control for a Mobile
Vehicle Disturbed by the Skid-Slip Phenomena, Journal
of Robotic and Intelligent Systems, vol. 59, issues 3-4, pp.
341-365, September 2010.
[8] . Lucet, C. Grand, D. Sall, P. Bidaud, Dynamic sliding
mode control of a four-wheel skid-steering vehicle in
presence of sliding, Proc. of the 17th CISM-IFToMM
Symposium on Robot Design, Dynamics, and Control,
Tokyo, Japan, July 2008.