Adaptive vehicle skid control

E. Faruk Kececi
a,
*
, Gang Tao
b
a
Department of Mechanical Engineering, Izmir Institute of Technology, Urla, Izmir 35430, Turkey
b
Department of Electrical and Computer Engineering, University of Virginia, Charlottesville, VA 22903, USA
Received 21 September 2004; accepted 15 October 2005
Abstract
In this paper, adaptive vehicle skid control, for stability and tracking of a vehicle during slippage of its wheels without braking, is
addressed. Two adaptive control algorithms are developed: one for the case when no road condition information is available, and
one for the case when certain information is known only about the instant type of road surface on which the vehicle is moving. The
vehicle control system with an adaptive control law keeps the speed of the vehicle as desired by applying more power to the drive wheels
where the additional driving force at the non-skidding wheel will compensate for the loss of the driving force at the skidding wheel, and
also arranges the direction of the vehicle motion by changing the steering angle of the two front steering wheels. Stability analysis proves
that the vehicle position and velocity errors are both bounded. With additional road surface information available, the adaptive control
system guarantees that the vehicle position error and velocity error converge to zero asymptotically even if the road surface parameters
are unknown.
2006 Elsevier Ltd. All rights reserved.
Keywords: Active steering control; Adaptive vehicle skid control; Driver assistance system; Non-linear adaptive control design; Vehicle stability and
tracking
1. Introduction
Initially, active control of vehicle dynamic systems is
used to reduce fuel consumption and emissions [1]. In cur-
rent automotive technology, active control is also used to
increase the safety and the reliability of the vehicle ride
and handling. Special control systems are designed for dif-
ferent parts of the vehicle dynamics [24]. Anti-lock brake
systems (ABS) are designed to prevent skidding of the
wheels during braking [5,6], while traction control systems
(TCS) are accomplishing the same objective, preventing the
slipping of the wheels, during acceleration [6]. When the
vehicle is on a slippery surface, because of the drop at
the coecient of road adhesion, the drive wheels may slip.
The traction control system reduces the engine torque or
applies brakes to the slipping wheels and brings the slip-
ping wheels into the desirable skid range. In order to
increase the stability of the vehicle during cornering, active
yaw control systems are designed, where either by braking
or by transferring the torque between the wheels, the speed
of the vehicle is decreased to critical speed to turn the cor-
ner and yaw control is achieved [7,9,10]. Vehicle state esti-
mation [11,12] and road condition estimation [13] methods
are used to improve the performance of ABS, TCS and
active yaw control systems. In [1416] controller algo-
rithms are designed to compensate for the error in the
lateral and yaw motion where sliding mode control and
minimization of cost function are used.
In order to increase the vehicle handling performance,
advanced hardware designs are also used. In four-wheel
steering vehicles, the rear wheels can also be steered
[6,17] which increases the steering ability of the vehicle.
The steer by wire mechanism which has replaced the con-
ventional steering mechanism with electrical motors,
rotates the steering wheels (which are the two front wheels
0957-4158/$ - see front matter 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.mechatronics.2005.10.005
*
Corresponding author. Tel.: +90 232 750 6618; fax: +90 232 750 6515.
E-mail addresses: [email protected] (E.F. Kececi), gt9s@virgi-
nia.edu (G. Tao).
URL: http://robotics.iyte.edu.tr (E.F. Kececi).
Mechatronics 16 (2006) 291301
of the vehicle) according to the drivers intentions [1820].
This system gives the ability to implement the automatic
steering concepts, which are used in driver assistance
systems.
Direct yaw moment control systems, driver assistance
systems, for vehicle handling at the high speed steering,
are employed to stabilize the motion of the vehicle by brak-
ing the slipping wheel and bringing it to a skid range so
that it can start applying driving force again [21]. These
systems are used in applications such as lane change [14]
and lane keeping [22].
In existing vehicle skid control systems, the skidding
wheel is detected and the power applied to this wheel is
reduced until the traction is regained. However, when the
total vehicle dynamic is considered, after the skidding
occurs, in order to regain control of the vehicle, the total
speed of the vehicle is reduced, which can be a problem
in trac when it causes some vehicles to come to a sudden
stop.
When a vehicle with an existing skid control technology
is considered, where the total speed of the vehicle is
decreased to regain traction at the skidding wheel, if the
vehicles behind this slowing down vehicle do not need to
slow down, because they do not need to have any traction
problem, they can hit the slowing down vehicle. The big-
gest advantage of the proposed design is that it does not
require for the vehicle to decrease its speed, so the vehicle
will not cause any danger to the other vehicles in the trac.
In this paper, the concept of controlling the stability of
the vehicle during slippage of the wheels without braking,
which can be applied to electrical vehicles [7] as well as
the all wheel drive vehicles, where the drive wheels and
steering wheels can be controlled separately, has been
developed. An adaptive control algorithm is developed
for the case when no road condition information is avail-
able. Stability analysis proves that the vehicle position
and velocity errors are both bounded. If the additional
road surface information available, the adaptive control
system guarantees that the vehicle position error and veloc-
ity error converge to zero asymptotically even if the road
surface parameters are unknown.
This research is the design of adaptive fault tolerant con-
trol algorithm. The application is the stability and tracking
control of a vehicle and the failure is described as the slip-
page of the wheels of the vehicle. The control algorithm is
designed to overcome the faults arising from the slippage
and to keep the vehicle at the desired speed and direction
intended by the driver.
Another application of this control algorithm is autono-
mous vehicles in a platoon system, where the following
vehicles should stay at the same speed, acceleration, and
direction with the lead vehicle. Preventing the slippage of
the wheels of the both the leading and the following vehicle
can stop any error at the speed acceleration and direction,
and as a consequence any possible collusion.
This paper is organized as follows: The driver assistance
systems are introduced and the new concept, adaptive vehi-
cle skid control, is explained in Section 1. System structure
of an electrical vehicle is studied in Section 2 and dynamic
equations of the system are formulated. In Section 3 an
adaptive control approach is used to compensate the un-
certainties arising from the road conditions. Section 3.1
explains the operation of the system and in Section 3.2
the problem, changing road friction coecient, is modeled.
In Section 3.3 adaptive control schemes are developed
for the unknown and known road condition cases and
with Lyapunov stability analyses, the boundedness of the
closed-loop signals are proved. Simulation results show
the eectiveness of the control algorithm in Section 4.
Conclusions and future work for this research are given
in Section 5.
2. System structure and dynamics
In order to demonstrate the adaptive controller design
for the wheel slip/skid problem, an electrical vehicle is
designed, which is shown in Fig. 1. The back wheels of
the vehicle have in-wheel electrical motors and the speed
of the back wheels can be controlled separately. The front
wheels are used for the purpose of steering only and are
accompanied by steer by wire system.
For low speed applications, kinematic steering (a.k.a.
Ackermann steering) is used to model the motion of the
vehicle where the sideslip angles are small so that, the
wheels are assumed to be pure rolling. A single-track model
(a.k.a. bicycle model) is used to simplify the analysis. The
wheels on an axle are represented by a single wheel with
the double of cornering stiness.
In this new study, dynamic steering is used, considering
the sideslip angles are big enough to create cornering
forces. A single-tack model is not applicable since the back
wheels are controlled separately and the cornering forces
eecting each front and back wheel can be dierent because
of the possible dierent road conditions acting on each
wheel.
When the model shown in Fig. 2 is considered, the fol-
lowing symbols are utilized as: V is the velocity vector; u
is the longitudinal velocity; v is the lateral velocity; b is
the sideslip angle; r is the yaw rate; CM is the center of
mass; l
f
and l
r
are the distances between the CM location
and the front and the back axles respectively; a
r
is the back
wheel slip angle; l
t
is the wheel tread; F
dl
is the driving force
applied by the left back wheel; F
dr
is the driving force
applied by the right back wheel; F
bl
is the lateral force at
the left back wheel; F
br
is the lateral force at the right back
wheel; F

is the lateral force at the left front wheel; F

fr
is the
lateral force at the right front wheel; a

is the left front

wheel slip angle; d
l
is the left front wheel steering angle;
a
fr
is the right front wheel slip angle and d
r
is the right front
wheel steering angle.
Moreover, the vehicle is assumed to be a rigid body
moving on a at surface and the vehicle has back wheel
drive and the front wheels are free rolling. It is also
assumed that air resistance, load transfer between the axles,
292 E.F. Kececi, G. Tao / Mechatronics 16 (2006) 291301
suspension system dynamics, caster eect and tire dynamics
are not included in the dynamics of the vehicle. Interested
readers can nd more detailed analysis of the vehicle
dynamics in [8], since this research is more focused on
the design of an adaptive controller, these simplied
dynamic equations will be used. With these assumptions,
dynamic equations of the system are formulated as
m _ u rv F
dl
F
dr
F
fl
sind
l
F
fr
sind
r
2:1
m_ v ru F
bl
F
br
F
fl
cosd
l
F
fr
cosd
r
2:2
J _ r l
f
F
fl
sind
l
F
fr
sind
r
l
r
F
bl
F
br
l
t
F
dr
F
dl

2:3
where m is the vehicle mass and J is the moment of inertia
about the vehicle mass center. The rst and second equa-
tions show the longitudinal and lateral dynamics respec-
tively, where the third equation formulates the yaw
motion dynamics. Since the roll and pitch motions are
slower than yaw motion [23], they are not included into
the dynamic equations.
From the drive wheel dynamics, the driving forces, F
dl
and F
dr
, are formulated as
F
dj

s
j
R
if
s
j
R
6
W
j
l
bj
r
W
j
l
bj
r
if
s
j
R
>
W
j
l
bj
r
_

_
2:4
where b indicates the back axle, j indicates the wheel, left or
right, s
j
is the driving torque applied to the wheel and R is
the wheel radius, W
j
is the vehicle weight on the wheel
l
bj
(r) is the road surface friction coecient which is a func-
tion of r wheel slip ratio.
The maximum driving force can be applied to the wheel
is formulated as
W j
l
bj
r
and if the applied driving force value
Fig. 2. Dynamical steering of the vehicle.
Fig. 1. Prototype electrical vehicle.
E.F. Kececi, G. Tao / Mechatronics 16 (2006) 291301 293
sj
R
is bigger than the maximum applicable force, the driving
force will be equal to the maximum value.
The lateral force eecting each wheel is formulated as
the function of the cornering stiness C
ij
of the wheel
and sideslip angle a
ij
as
F
ij
C
ij
a
ij
2:5
where i indicates the axle: i = f for front or i = b for
back, j indicates the wheel: j = l for left or j = r for
right.
Cornering stiness C
ij
is a function of the road surface
friction coecient and formulated as
C
ij
Cl
ij
2:6
where C is the cornering coecient of the tire and l
ij
is the
friction coecient eecting the jth wheel at the ith axle.
Sideslip angle a
ij
is formulated as
a
ij
arctan
v rl
i
u rl
t
_ _
d
ij
2:7
This study assumes equal slip angles on the left and right
wheels and only the front wheels are steered. For the sim-
plication of the system steering angles at the front right
and left wheels, d
fr
and d

, are assumed to be the same

steering angle d = d
fr
= d

. Linear tire model is used and

the cornering stiness coecient C is assumed to be the
same for each tire. It should also be noted that the sideslip
occurs only when the longitudinal motion exists, ensuring
that cornering forces are nite. Under these assumptions,
the cornering forces are formulated as
F
bl
Cl
bl
arctan
v rl
r
u rl
t
_ _
2:8
F
br
Cl
br
arctan
v rl
r
u rl
t
_ _
2:9
F
fl
Cl
fl
arctan
v rl
f
u rl
t
_ _
d
_ _
2:10
F
fr
Cl
fr
arctan
v rl
f
u rl
t
_ _
d
_ _
2:11
Substituting Eqs. (2.8)(2.11) into (2.1)(2.3), the
dynamic equations of the system are found as
m_ u mrv
s
r
R

s
l
R
C sind arctan
v rl
f
u rl
t
_ _
d
_ _
l
fl
l
fr
2:12
m_ v mru C arctan
v rl
r
u rl
t
_ _
l
bl
l
br

C cosd arctan
v rl
f
u rl
t
_ _
d
_ _
l
fl
l
fr
2:13
J _ r l
f
C sind arctan
v rl
f
u rl
t
_ _
d
_ _
l
fl
l
fr

l
r
C arctan
v rl
r
u rl
t
_ _
l
bl
l
br

l
t
R
s
r
s
l

2:14
which can be rewritten as
Dq f _ q; l
ij
g _ q; s; l
ij
2:15
where l
ij
is the friction coecient eecting the jth wheel at
the ith axle, i indicates the axle, front or back, j indicates
the wheel, left or right, the sate vector _ q and the control
input vector s are formulated as respectively, _ q u; v; r
T
and s = [d, s
r
, s
l
]. The positive-denite mass matrix D, the
term containing the unaected dynamics from the control
input f _ q; l
ij
and the term contains the controllable
dynamics g _ q; s; l
ij
are formulated as respectively
D
m 0 0
0 m 0
0 0 J
_

_
_

_ 2:16
f _ q; l
ij

mrv
mruCarctan
v rl
r
url
t
_ _
l
bl
l
br

l
r
Carctan
vrl
r
url
t
_ _
l
bl
l
br

_
_

_
2:17
g_ q; s; l
ij

Csind arctan
v rl
f
url
t
_ _
d
_ _
l
fl
l
fr

s
r
R

s
l
R
Ccosd arctan
vrl
f
url
t
_ _
d
_ _
l
fl
l
fr

l
f
Csind arctan
v rl
f
url
t
_ _
d
_ _
l
fl
l
fr

l
t
R
s
r
s
l

_
_

_
2:18
3. Adaptive wheel skid control
3.1. System operation
Driver assistance systems can be categorized into three
dierent groups by their application area [4]. Longitudinal
control involves keeping the vehicle at a designated dis-
tance from the prior vehicle maintaining a relatively con-
stant speed. This is accomplished by using cruise control
and adaptive cruise control systems, where radar is added
to a standard cruise control to measure the distance and
the speed of the prior vehicle to adjust the following dis-
tance and the speed of the vehicle. Anti-lock brake systems
(ABS) and traction control systems (TCS) can be cate-
gorized in the longitudinal vehicle control systems, where
these systems are designed to prevent skidding in the longi-
tudinal dynamics.
Lateral control of the vehicle is achieved by look-down
and look-ahead reference systems and accomplishes the
low speed steering of the vehicle. Lane keeping, lane chang-
ing, turning, avoiding obstacles, steering over correction
because of a at tire and lane departure problems are
solved by lateral control algorithms.
The direct yaw moment control system, a combination
of longitudinal and lateral control systems, is used to con-
trol the vehicle at high speed steering. This system stabilizes
294 E.F. Kececi, G. Tao / Mechatronics 16 (2006) 291301
the motion of the vehicle by braking the slipping wheel and
bringing it to a skid range that it can start applying driving
force again [21]. In the current direct yaw moment system
technology, the skidding wheel is detected and the power
applied to this wheel is reduced, until the traction is
regained. However, when the total vehicle dynamic is con-
sidered, after the skidding occurs, in order to regain control
of the vehicle, the total speed of the vehicle is reduced,
which can be a problem in trac, because accidents occur
when any vehicles speed suddenly decreases.
In the proposed system, the speed and the direction of
the vehicle are designed to be maintained as desired by
the driver. A motion stabilizer system whose block diagram
is shown in Fig. 3 is proposed. The operation of the motion
stabilizer is as follows: The speed and the direction of the
vehicle are determined by the driver. In case of a wheel slip-
page, an adaptive control algorithm keeps the speed of the
vehicle as desired by applying more power to the drive
wheels where the additional driving force at the non-skid-
ding wheel will compensate for the loss of the driving force
at the skidding wheel, and the controller also arranges the
direction of the vehicle by changing the steering angle of
the steering wheels.
3.2. Problem formulation
While a ground vehicle is travelling on the road, the
forces acting on the vehicle are the functions of the road
friction coecient and these driving forces are formulated
as
F
dj

s
j
R
if
s
j
R
6
W
j
l
bj
r
W
j
l
bj
r
if
s
j
R
>
W
j
l
bj
r
_

_
3:1
where b indicates the back axle, j indicates the wheel, left or
right, s
j
is the driving torque applied to the wheel and R is
the wheel radius, W
j
is the vehicle weight on the wheel
l
bj
(r) is the road surface friction coecient which is a func-
tion of r wheel slip ratio. Lateral forces are formulated as
F
ij
Cl
ij
a
ij
3:2
where C is the cornering coecient of the tire and l
ij
is the
friction coecient, a
ij
is the sideslip angle eecting the jth
wheel at the ith axle.
On dierent road surfaces, the road friction coecients,
which are functions of slip ratio, vary and for an example
tire are shown in Fig. 4. The road friction coecient is
found experimentally, but also can be approximated by
an analytical expression. A general expression in the range
of 1 < r < 1 is
l a1 e
br
cr
2
dr 3:3
where the coecients a, b, c and d are obtained from the
experimental curves and r is the slip ratio and is dened as
Fig. 3. Block diagram of the proposed driver assistance system.
0 0.5 1
0
0.4
0.8
1.2
Dry
Wet
Snow
Ice
Fig. 4. Tire friction coecient vs. wheel slip.
E.F. Kececi, G. Tao / Mechatronics 16 (2006) 291301 295
r
Rx v
Rx
3:4
where R is the wheel radius, x is the angular speed of the
wheel and v is the linear speed of the wheel center. The
ABS system will not allow the wheel to get locked, this
means x will not be zero and r, slip ratio, is always dened.
At dierent surfaces (i.e., dry asphalt, wet asphalt or ice)
the a, b, c, d coecients will dier. The uncertainty in the
road friction confection is expressed as
l
ij
l
ij1
k
ij1
t l
ij2
k
ij2
t l
ijk
k
ijk
t 3:5
where l
ij
is the actual road friction coecient at the ith axle
jth wheel, l
ijp
, p = 1, . . . , k, are road friction coecients for
dierent surfaces, which last a certain period of time, and
k
ijp
(t) are time functions indicating which values of l
ijp
are taken by l
ij
at a given time. l
ij
is a function of time
showing the ith axle jth wheel road friction coecient.
k
ijp

1 if l
ij
l
ijp
0 otherwise
_
3:6
Function k
ijp
(t) indicates only one value can be active at
any time, that is k
ijm
(t)k
ijn
(t) = 0, for m 5n, and k
ij1
+
k
ij2
+ + k
ijk
= 1.
The following two cases are possible for k
ijp
(t)
(i) The road condition information is not available, that
is, k
ijp
(t) are unknown;
(ii) The road condition is measured, that is, k
ijp
(t) are
known.
While the real values of l
ijp
are not known, these bounds
can be found experimentally. When the road condition is
measured, k
ijp
(t) are known.
During the eld test the friction coecients will require
tuning, because there is no exact friction coecient value
for an icy road. However, the adaptive control algorithm
ensures that with the parameter update laws, arbitrarily
guessed unknown parameter values are converged to the
values that makes the system error converge to zero, where
the system is stable.
The controller objective is to design a feedback control
law s for the system (2.15) to ensure that all closed-loop
system signals and parameter estimates are bounded, and
that the system output q(t) asymptotically tracks a given
reference output q
d
(t). In other words, the controller objec-
tive is to keep the speed and the direction of the vehicle as
desired by the driver, despite the uncertainties arising from
the road conditions.
3.3. Adaptive controller design
Because of computing diculties there were some
assumptions made in the previous section during the for-
mulation of the mathematical model of the system. There
are also measuring diculties in the system, which causes
more assumptions, such as the mass and the inertia of
the vehicle, since these values would change with the num-
ber of passengers. These assumptions cause uncertainties
and in order to compensate for these uncertainties in the
system, adaptive control methods are used. The adaptive
control system uses the knowledge from plant dynamics
and updates the control parameters on-line to eliminate
the tracking error.
3.3.1. Design with unknown road condition
When there is no information available about the road
condition, the adaptive control scheme can be derived in
the following procedure. The term, including the control
inputs in the dynamical equations of the system,
g _ q; s; l
ij
in (2.18), is expressed as
g _ q; s; l
ij
K
s
s 3:7
where s = [d, s
r
, s
l
]
T
and
K
s

C sind A d l
fl
l
fr

d
1
R
1
R
C cosd A d l
fl
l
fr

d
0 0
l
f
C sind A d l
fl
l
fr

d
l
t
R

l
t
R
_

_
_

_
2 R
33
3:8
with A arctan
vrl
f
urlt
.
Let
^
K
s
be the estimate of K
s
and choose the control law
as
s
^
K
1
s
s
d
3:9
where s
d
is a feedback control law designed based on the
ideal system
Dq f _ q; l
ij
s
d
3:10
The existence of
^
K
1
s
is to be ensured by a projection
algorithm in the adaptive law for updating the estimate
^
K
s
t. From (2.15), (3.7) and (3.9), the closed-loop dynamic
equation is
Dq f _ q; l
ij
K
s

^
K
s
s s
d
3:11
The ideal matching parameter for
^
K
s
t is K
s
.
To design an adaptive control scheme to handle param-
eter uncertainties, for _ q u; v; r
T
and _ q
d
u
d
; v
d
; r
d

T
, the
tracking error e and the ltered tracking errors s; _ s are
dened as
e q q
d
; _ e _ q _ q
d
3:12
s _ e ge; _ s e g_ e 3:13
where g 2 R
3 3
> 0 is a diagonal gain matrix.
At rst the ideal system is expressed as Dq f _ q; l
ij

s
d
as D_ s Y _ q; q; hh s
d
, where
Y _ q; q; hh Dh f _ q; l
ij
; h q
d
g_ e 3:14
where Y _ q; q; h 2 R
34
is a known matrix function,
h(m, l
ij
, J) 2 R
41
is an unknown parameter vector and
are formulated as
296 E.F. Kececi, G. Tao / Mechatronics 16 (2006) 291301
Y _ q; q; h
h
1
rv 0 0 0
h
2
ru 0 CA CA
0 h
3
l
r
CA l
r
CA
_

_
_

_ 3:15
h m; J; l
bl
; l
br

T
. 3:16
The actual system (3.11) is expressed as
D_ s Y h
^
K
s
K
s
s s
d
3:17
for which K
s
s can be further expressed as
K
s
s Y
s
h
s
3:18
where
h
s
l
fr
; l
fl
;
1
R
_ _
T
3:19
is the parameter vector, and
Y
s

CsindAd CsindAd s
r
s
l

CcosdAd CcosdAd 0
l
f
CsindAd l
f
CsindAd l
t
s
r
s
l

_
_

_
3:20
is the known matrix function.
Let
^
h
s
be the estimate of h
s
. Then, the dynamical system
(3.17) becomes
D_ s Y h Y
s

^
h
s
h
s
s
d
3:21
For this system, the control law is designed as
s
d
Y
^
h K
D
s 3:22
where K
D
K
T
D
> 0, K
D
2 R
33
is a gain matrix,
^
h is the
estimate of h, updated from an adaptive law. This control
law could also be used for the ideal system D_ s Y h s
d
.
With the control law (3.22), the closed-loop system is
D_ s Y
^
h h Y
s

^
h
s
h
s
K
D
s 3:23
A Lyapunov-like positive-denite scalar function is
chosen as
V
1
t
1
2
s
T
Ds
1
2
~
h
T
C
1
1
~
h
1
2
~
h
T
s
C
1
2
~
h
s
3:24
where
~
h
^
h h,
~
h
s

^
h
s
h
s
, C
1
C
T
1
> 0 and C
2

C
T
2
> 0.
For the time t in each interval when the values of h and
h
s
do not change, dierentiating (3.24) with respect to the
time t and using (3.23) yields
_
V
1

~
h
T
C
1
1
_
~
h Y
T
s
~
h
T
s
C
1
2
_
~
h
s
Y
T
s
s s
T
K
D
s 3:25
Choose the adaptive parameter update laws for
_
^
h and
_
^
h
s
as
_
^
h
_
~
h C
1
Y
T
s 3:26
_
^
h
s

_
~
h
s
C
2
Y
T
s
s 3:27
which are dened for all t P0. Then, the derivative of the
Lyapunov-like function is found as
_
V
1
s
T
tK
D
st 6 0 3:28
for t in each interval when the values of h and h
s
do not
change.
The Lyapunov-like function V
1
(t) is positive-denite
and
_
V
1
is negative semi-denite. Stability in the sense of
Lyapunov has been shown in that the position tracking
and velocity tracking errors are both bounded. When a slip
occurs at any of the wheels of the vehicle, a bounded jump
occurs in the Lyapunov-like function V
1
(t), where its deriv-
ative
_
V
1
is undened. The stability of the system is valid
for every each time interval between the occurrence of each
jump. Due to possible persistent uncertain jumps, the
tracking error may not converge to zero. However, if there
are only a nite number of jumps, the developed adaptive
control scheme ensures that the tracking error converges
to zero asymptotically. For the case when the parameter
jumps are persistent, that is, the vehicle keeps entering a
surface at one time and another surface at another time,
practical tracking is achievable if the time interval of
switching surface is large enough. In this case, to ensure
asymptotic tracking, a modied adaptive control scheme
can be developed, using certain additional system informa-
tion (see the known road condition case).
In order to ensure that
^
K
s
t has a bounded inverse
^
K
1
s
t, a projection algorithm is used in the design of the
adaptive update law. The term
^
K
s
s is expressed as
^
K
s
s Y
s
d; s
r
; s
l

^
h
s
3:29
where
^
h
s
is the estimate of the parameter vector and
Y
s
(d, s
r
, s
l
) is the known signal vector.
The adaptive update law for
^
h
s
is given in Eq. (3.27) as
_
^
h
s

_
~
h
s
C
2
Y
T
s
s 3:30
and by applying a projection algorithm dened as
_
^
h
s
Gt 3:31
Gt G
1
t; . . . ; G
4
t
T
2 R
4
3:32
where the size of G(t) is equal to the size of h
s
which con-
sists of four unknown parameters namely l
fr
, l

, l
br
and
l
bl
.
G
k
t
if
^
h
sk
t 2 h
a
sk
; h
b
sk
or
F
k
t if
^
h
sk
t h
a
sk
and F
k
t P0 or
if
^
h
sk
t h
b
sk
and F
k
t 6 0
0 otherwise
_

_
3:33
where F(t) is dened as F t F
1
t; . . . ; F
4
t
T
C
2
Y
T
s
s
and h
sk
2 h
a
sk
; h
b
sk
; k 1; . . . ; 4,
^
h
sk
and h
sk
are the kth com-
ponents of
^
h
s
and h
s
, respectively, h
a
sk
; h
b
sk
; k 1; . . . ; 4, are
known such that for any
^
h
sk
2 h
a
sk
; h
b
sk
, where h
a
sk
is the
upper limit of l
ij
, which is l
dry
and h
b
sk
is the lower limit
of l
ij
, which is l
ice
, k = 1, . . . , 4, the resulting
^
K
s
t is invert-
ible, that is,
^
K
1
s
t is bounded, and n is the degree of the
dierential equations representing the dynamics of the sys-
tem, which is two in this study. It can be veried that the
E.F. Kececi, G. Tao / Mechatronics 16 (2006) 291301 297
desired property (3.28) still holds, and in addition, it is
ensured that
^
h
sk
2 h
a
sk
; h
b
sk
; k 1; . . . ; 4, so that
^
K
1
s
t is
bounded.
3.3.2. Design with known road condition
It is possible to measure the road condition by using a
millimeter radiometer, a radar sensor [2426], in the sense
it is known whether the vehicle is on a dry surface or on
an icy surface, but the road friction coecients are
assumed to be unknown, that is, k
ijp
(t) are known, but l
ijp
are unknown.
Using (3.5) and (3.6), we express g _ q; s; l
ij
in (2.15) as
g _ q; s; l
ij
Ks
1
k
ij1
Ks
2
k
ij2
Ks
k
k
ijk
s 3:34
where
Ks
p

C sind A d l
flp
l
frp

d
1
R
1
R
C cosd A d l
flp
l
frp

d
0 0
l
f
C sind A d l
flp
l
frp

d
l
t
R

l
t
R
_

_
_

_
p 1; 2; . . . ; k
3:35
where A arctan
vrl
f
urlt
.
Let
^
K
sp
be the estimate of K
sp
, p = 1,2, . . . , k, and choose
the control law as
s
^
K
1
s1
k
ij1
t
^
K
1
sp
k
ijp
ts
d
3:36
Then Eq. (3.11) becomes
Dqf _ q; l
ij
K
s1

^
K
s1
k
ij1
t K
sp

^
K
sp
k
ijp
tss
d
.
3:37
Considering that s
p
= k
ijp
s and inserting this equation
into the previous equation results as
Dq f _ q; l
ij
K
s1

^
K
s1
s
1
K
sp

^
K
sp
s
k
s
d
. 3:38
f _ q; l
ij
formulated as
f _ q; l
ij
Y _ q; q; hh
1
k
ij1
h
2
k
ij2
h
k
k
ijk
3:39
By including Y
p
= Yk
ijp
and Eq. (3.39) into Eq. (3.38),
the actual system is expressed as
D_ s Y
1
h
1
Y
2
h
2
Y
k
h
k
K
s1

^
K
s1
s
1
K
s2

^
K
s2
s
2
K
sk

^
K
sk
s
k
s
d
3:40
Considering that
K
s
s Y
s1
d; s
r
; s
l
h
s1
Y
s2
d; s
r
; s
l
h
s2
Y
sk
d; s
r
; s
l
h
sk
3:41
^
K
s
s Y
s1
d; s
r
; s
l

^
h
s1
Y
s2
d; s
r
; s
l

^
h
s2
Y
sk
d; s
r
; s
l

^
h
sk
3:42
and by choosing the control law as
s
d
Y
1
^
h
1
Y
2
^
h
2
Y
k
^
h
k
K
D
s 3:43
the closed-loop system becomes
D_ s Y
1

^
h
1
h
1
Y
2

^
h
2
h
2
Y
k

^
h
k
h
k

Y
s1

^
h
s1
h
s1
Y
s2

^
h
s2
h
s2

Y
sk

^
h
sk
h
sk
K
D
s 3:44
A Lyapunov-like positive-denite scalar function is
chosen as
V
2

1
2
s
T
Ds
1
2

k
p1
~
h
T
p
C
1
p
~
h
p

1
2

k
p1
~
h
T
sp
C
1
s2
~
h
sp
3:45
For the time t in each interval when the values of h
p
and
h
sp
do not change, dierentiating (3.45) with respect to the
time t and using (3.44) yields
_
V
2

k
p1
~
h
T
p
C
1
p
_
~
h
p
Y
T
p
s

k
p1
~
h
T
sp
C
1
sp
_
~
h
sp
Y
T
sp
s s
T
K
D
s 3:46
Choose the adaptive parameter update laws for
_
^
h
p
and
_
^
h
sp
as
_
^
h
p

_
~
h
p
C
p
Y
T
p
s 3:47
_
^
h
sp

_
~
h
sp
C
2
Y
T
sp
s; p 1; 2; . . . ; k 3:48
which are dened for all t P0, and the derivative of the
Lyapunov-like function is found as
_
V
2
s
T
K
D
st 6 0 3:49
for t in each interval when the values of h
p
and h
sp
do not
change, the Lyapunov-like function V
2
(t) is valid over the
time interval for each change in the road friction coecient
for each wheel.
The fact that
_
V
2
s
T
K
D
s 6 0 implies s 2 L
2
\ L
1
and
h
p
and h
sp
2 L
1
for p = 1, . . . , k. Since s = _ e + ge, it implies
e and _ e 2 L
2
\ L
1
, and also q and _ q 2 L
1
. From (3.22) it
follows that s
d
and _ s 2 L
1
which results e 2 L
1
. By apply-
ing the Barbalat Lemma [27] it is concluded that the posi-
tion tracking error e and velocity tracking error _ e go to
zero as t goes to 1.
In summary, it is proven that with the control law (3.43)
and adaptive update laws (3.47) and (3.48), when the road
condition is measured, it is guaranteed that all closed-loop
signals are bounded and the tracking error e(t) and _ e(t) go
to zero as t goes to 1.
In order to ensure that
^
K
s
t has a bounded inverse
^
K
1
s
t, a projection algorithm is used in the design of the
adaptive update law same as before.
4. Simulation results
The vehicle model formulated in Eqs. (2.15)(2.18) is
used in the simulations. Since, this study is more focused
298 E.F. Kececi, G. Tao / Mechatronics 16 (2006) 291301
on design of the control algorithm, the assumptions men-
tioned in Chapter 2 are made. Another assumption made
in this research is also the fact that the drivers reaction
is not important, since the purpose of this research is to
demonstrate the design of the control algorithm. In reality,
the drivers reaction will be very important, and should be
included in the eld test of this algorithm on a model car.
This is what we would like to achieve in the future research.
In this simulation study, the vehicle design shown in
Fig. 1 is driven on a straight road with a 10 m/s constant
0 2 4 6 8 10 12 14 16 18 20
0
2
4
X
0 2 4 6 8 10 12 14 16 18 20
1
0
0.1
0.2
0.3
Y
0 2 4 6 8 10 12 14 16 18 20
2
1
0
0.1
0.2
Time (sec)
Fig. 5. Position errors.
0 2 4 6 8 10 12 14 16 18 20
0
20
40
0 2 4 6 8 10 12 14 16 18 20
0
2000
4000
R
0 2 4 6 8 10 12 14 16 18 20
0
2000
4000
L
Time (sec)
Fig. 6. Control inputs.
E.F. Kececi, G. Tao / Mechatronics 16 (2006) 291301 299
speed. At the fth second the driver realizes the ice on the
road and maneuvers the vehicle to the left, but the rear
right wheel drives on the ice and the eect of the changing
friction coecient is shown in the following gures. Fig. 5
shows the position errors in the longitudinal, lateral and
angular directions. Fig. 6 demonstrates the change in the
control inputs because of the changing road conditions.
The eect of the known road condition controller is shown
0 20 40 60 80 100 120 140 160 180 200
1
0
1
2
3
4
D
e
s
i
r
e
d
0 20 40 60 80 100 120 140 160 180 200
1
0
1
2
3
4
W
i
t
h

c
o
n
t
r
o
l
l
e
r
1.5 1 0.5 0 0.5 1
x 10
9
5
0
5
10
x 10
11
Lateral vs. longitudinal position
W
i
t
h
o
u
t

c
o
n
t
r
o
l
l
e
r
Fig. 7. Eect of the controller.
0 2 4 6 8 10 12 14 16 18 20
0
0.5
0.5
1
1.5
AVSC vs. ESP
AVSC
ESP
0 2 4 6 8 10 12 14 16 18 20
0
1
1
2
3
L
a
t
e
r
a
l

c
h
a
n
g
e

(
m
)
0 2 4 6 8 10 12 14 16 18 20
0
50
100
150
200
Time (sec)
L
o
n
g
i
t
u
d
i
n
a
l

c
h
a
n
g
e

(
m
)
Fig. 8. Comparison of the AVSC system to an ESP system.
300 E.F. Kececi, G. Tao / Mechatronics 16 (2006) 291301
in Fig. 7, where with the controller can achieve the desired
performance, but without the controller the system
becomes unstable.
Fig. 8. shows the advantages of the proposed control
system Adaptive Vehicle Skid Control (AVSC) to an ESP
system studied in [28], where the earlier simulation is
repeated for both AVSC and ESP systems. The result is
the vehicle with an AVSC system continues with the same
speed whereas the vehicle with an ESP system cannot keep
its intended position in the longitudinal direction.
5. Concluding remarks
In this research, adaptive vehicle skid control concept is
developed. Dynamic equations of an electric vehicle are for-
mulated and the eect of the road friction coecient is stud-
ied. When the road condition information is not available,
the boundedness of the position and velocity error is
proved. In the second part of the research, the performance
of the controller is examined when the road condition is
measured (but the road friction coecients are assumed
not to be known). Simulation results show the eectiveness
of the controller schemes. The eect of a at tire and roll-
over resistance of the vehicle, by selecting the steering angles
d

and d
fr
dierently, will also be studied for future research.
References
[1] van der Voort MC. FEST. A new driver support tool that reduces fuel
consumption and emissions. Int Conf Adv Driver Assistance Syst
2001:903.
[2] Mammar S, Baghdassarian VB. Two-degree-of-freedom formulation
of vehicle handling improvement by active steering. Proc Amer Contr
Conf 2000;1:1059.
[3] Fodor M, Yester J, Hrovat D. Active control of vehicle dynamics.
Proc Digital Avionics Syst Conf 1998;2:I14/18.
[4] Nobe SA, Fei-Yue W. An overview of recent developments in
automated lateral and longitudinal vehicle controls. Int Conf Syst,
Man, Cyber 2001;5:344752.
[5] Johansen TA, Kalkkuhl J, Ludemann J, Petersen I. Hybrid control
strategies in ABS. Proc Amer Contr Conf 2001;2:17045.
[6] Wong JY. Theory of ground vehicles. New York: John Wiley &
Sons; 2001.
[7] Sakai S, Sado H, Hori Y. Motion control in an electric vehicle with
four independently driven in-wheel motors. IEEE Trans Mech
1999;4(1):916.
[8] Genta G. Motor vehicle dynamics modeling and simulation. Singa-
pore: World Scientic Publishing Co. Pte. Ltd.; 1997.
[9] Drakunov SV, Ashra B, Rosiglioni A. Yaw control algorithm via
sliding mode control. Proc Amer Contr Conf 2000;1:5803.
[10] Kaoru S, Yoshiaki S. Application of active yaw control to vehicle
dynamics by utilizing driving/breaking force. JSAE Rev
1999;20(2):28995.
[11] Bevly DM, Gerdes JC, Wilson C, Gengsheng Z. The use of GPS
based velocity measurements for improved vehicle state estimation.
Proc Amer Contr Conf 2001;4:253842.
[12] Tseng HE, Ashra B, Madau D, Allen Brown T, Recker D. The
development of vehicle stability control at Ford. IEEE Trans Mech
1999;4(3):22334.
[13] Sado H, Sakai S, Hori Y. Road condition estimation for traction
control in electric vehicle. Int Symp Ind Electron 1999;2:9738.
[14] Mokhiamar O, Abe M. Active wheel steering and yaw moment
control combination to maximize stability as well as vehicle respon-
siveness during quick lane change for active vehicle handling safety.
J Automobile Eng 2002;216(2):11524.
[15] Matsumoto N, Kuraoka H, Ohba M. An experimental study on
vehicle lateral and yaw motion control. Int Conf Indust Electron,
Contr Instrum 1991;1:1138.
[16] Langson W, Alleyne A. Multivariable bilinear vehicle control using
steering and individual wheel torques. Proc Amer Contr Conf
1997;2:113640.
[17] Raksincharoensak P, Nagai M, Mouri H. Investigation of automatic
path tracking control using four-wheel steering vehicle. Proc Vehicle
Electron Conf 2001:737.
[18] Hayama R, Nishizaki K, Nakano S, Katou K. The vehicle stability
control responsibility improvement using steer-by-wire. Proc Intell
Vehicles Symp 2000:596601.
[19] Fukao T, Miyasaka S, Mori K, Adachi N, Osuka K. Active steering
systems based on model reference adaptive non-linear control. Proc
Intell Transport Syst 2001:5027.
[20] Mammar S, Sainte-Marie J, Glaser S. On the use of steer-by-wire
systems in lateral driving assistance applications. Int Workshop
Robot Human Interactive Commun 2001:48792.
[21] Yoshimoto K, Tanaka H, Kawakami S. Proposal of driver assistance
system for recovering vehicle stability from unstable states by
automatic steering. Int Vehicle Electron Conf 1999:5149.
[22] Sparbert J, Dietmayer K, Streller D. Lane detection and street type
classication using laser range images. Proc Intell Transport Syst
2001:4549.
[23] Kiencke U, Nielsen L. Automotive control systems. Singapore: SAE
International; 2000.
[24] Macelloni G, Ruisi R, Pampaloni P, Paloscia S. Microwave radiom-
etry for detecting road ice. Proc Geosci Remote Sensing Symp
1999;2:8913.
[25] Wollitzer M, Buechler J, Luy JF, Siart U, Schmidhammer E,
Detlefsen J, et al. Multifunctional radar sensor for automotive
application. IEEE Trans Microwave Theory Tech 1998;46(5):
7018.
[26] Kees N, Detlefsen J. Road surface classication by using a polari-
metric coherent radar module at millimeter waves. Proc Telesyst Conf
1994:958.
[27] Lewis FL, Abdallah CT, Dawson DM. Control of robot manipula-
tors. New York: Macmillan Publishing Company; 1993.
[28] Mokhiamar O, Abe M. How the four wheels should share forces in
an optimum cooperative chassis control. Contr Eng Practice
2006;14(3):295304.
E.F. Kececi, G. Tao / Mechatronics 16 (2006) 291301 301