Trajectory Tracking Control of Nonholonomic
Wheeled Mobile Robots
Combined Direct and Indirect Adaptive Control using Multiple Models Approach
Altan Onat
1
and Metin Ozkan
2
1
Electrical & Electronics Engineering Department, Anadolu University, Iki Eylul Kampusu, Eskisehir, Turkey
2
Computer Engineering Department, Eskisehir Osmangazi University, Bati Meselik Kampusu, Eskisehir, Turkey
Keywords: Combined Direct and Indirect Adaptive Control, Trajectory Tracking Control, Nonholonomic Wheeled
Mobile Robots, Multiple Models Approach.
Abstract: This paper presents a novel methodology for the trajectory tracking control of nonholonomic wheeled
mobile robots using multiple identification models. The overall control system includes two stages. In the
first stage, a kinematic controller developed by using kinematic model provides the required linear and
angular velocities of the robot for tracking a reference trajectory. In the second stage, the required velocities
are taken as the inputs to an adaptive dynamic controller which uses multiple adaptive models for the
parameter identification. The proposed adaptive dynamic controller is developed using a combined direct
and indirect adaptive control approach where both prediction and tracking errors are used for identification.
Simulation results show the effectiveness of the proposed combined direct and indirect control scheme and
multiple models approach.
1 INTRODUCTION
Tracking control of a wheeled mobile robot (WMR)
is one of the most attractive research areas for the
several decades. Many WMR models and control
schemes have been presented. Generally, the aim of
such schemes is either to utilize a kinematic
trajectory tracking controller or to construct and
integrate kinematic and dynamic controllers to track
a desired trajectory. Yutaka et al. (1990) proposed a
control rule to determine reasonable linear and
rotational velocities for a stable tracking control. An
integrated kinematic controller and a torque
controller with a dynamic extension for a
nonholonomic mobile robot have been presented by
Fierro and Lewis (1995). Yun and Yamamoto
(1992) have studied feedback linearization of a
WMR and its dynamic system. A complete dynamic
model of a WMR which makes it suitable to
consider rotational and translational velocities as
control signals has been given by De La Cruz and
Carelli (2006).
For the tracking control of a WMR, there are also
adaptive control frameworks in literature. Felipe et
al. (2008) have proposed an adaptive controller to
guide a WMR during trajectory tracking. In this
study reference velocities are generated using a
kinematic model, and then these values are
processed to compensate for the robot dynamics. An
adaptive trajectory tracking controller for a
nonholonomic WMR with a nonlinear control law
based on input-output feedback linearization has
been proposed by Khoshnam et al. (2010). Cao et al.
(2011) has proposed an adaptive kinematic
controller to generate the command of velocity
based on backstepping method, and then Zhengcai et
al. (2011) has proposed adopting the reference
model with a dynamic adaptive controller. Similarly,
a new kinematic adaptive controller integrated with
a torque controller for the dynamic model of a
nonholonomic WMR has been proposed by
Takanori et al.(2000). Pourboghrat and Karlsson
(2002) has used adaptive control rules for the
dynamics level of nonholonomic WMRs with
unknown dynamic parameters and a fixed posture
backstepping technique for tracking a reference
trajectory and stabilization. Petrov (2010) has
proposed an adaptive dynamic based path control for
a differential drive mobile robot.
The studies previously mentioned provide the
schemes of trajectory tracking, but they did not
95
Onat A. and Ozkan M..
Trajectory Tracking Control of Nonholonomic Wheeled Mobile Robots - Combined Direct and Indirect Adaptive Control using Multiple Models Approach.
DOI: 10.5220/0004039800950104
In Proceedings of the 9th International Conference on Informatics in Control, Automation and Robotics (ICINCO-2012), pages 95-104
ISBN: 978-989-8565-22-8
Copyright
c
2012 SCITEPRESS (Science and Technology Publications, Lda.)
focus on the transient behaviour. However, when the
parameter errors are very large, the transient
response of the system may include unacceptably
large peaks. Although the system is asymptotically
stable, the adaptive control approach may be in
applicable for some systems due to the transient
peaks. To overcome this difficulty, the enhancement
of the transient response using multiple models and
switching has been proposed for the linear systems
by Narendra and Balakrishnan (1997). Some
approaches using multiple models and switching for
nonlinear systems have been presented in several
studies. Narendra and George (2002) have presented
a multiple model, switching and tuning methodology
which improves the transient performance for a class
of nonlinear systems. A novel approach which
makes use of multiple identification models and
switching based on direct adaptive control scheme
has been proposed by Cezayirli and Ciliz (2007).
Besides composite approach where both prediction
and tracking errors are used in a combined direct and
indirect adaptive control framework has been
studied by (Ciliz and Narendra, 1995) and (Ciliz and
Cezayirli, 2004). Ye (2008) has proposed a multiple
model adaptive controller for nonlinear systems in
parametric-strict-feedback form. An adaptive control
of a class of single-input single-output (SISO)
nonlinear systems considering transient performance
improvement by using multiple models and
switching has been considered by Cezayirli and Ciliz
(2006 and 2008). Ciliz and Narendra (1994), Ciliz
and Tuncay (2005) have used a scheme consisting of
multiple models, switching and tuning for the
adaptive control of robotic manipulators.
Figure 1: Nonholonomic WMR.
The purpose of this paper is to present an
integrated kinematic and dynamic controller for the
trajectory tracking of a WMR that includes
parametric uncertainties in the dynamics. A
composite approach, in which both prediction and
tracking errors are used in a combined direct and
indirect adaptive control framework with multiple
identification models and switching, is used. There
are a few works which make use of the multiple
models approach for the control of the WMRs. De
La Cruz et al. (2008) has proposed a switching
control for a novel tracking adaptive control of
WMRs. Another method that uses multiple models
of the robot for its identification in an adaptive and
learning control framework has been presented by
D’Amico et al. (2006).
2 KINEMATICS AND DYNAMICS
Consider the WMR model given by (1). The
parameters are given in Table 1 and the system is
shown in Figure 1. The system is subjected to m
constraints:
() (,) () ()
T
M
qq Cqqq Bq A q
τλ
+=+
&& & &
(1)
where
n
qR
is generalized coordinates,
r
R
τ
is
the input vector,
m
R
λ
is the vector of constraint
forces,
()
nn
M
qR
×
is a symmetric positive-definite
inertia matrix,
(,)
nn
Cqq R
×
&
is coriolis matrix,
()
nr
B
qR
×
is the input transformation matrix, and
()
mn
A
qR
×
is the matrix associated with the
constraints.
Table 1: Model Parameters of WMR.
Parameter Description
r Driving wheel radius
2b Distance between two wheels
d Distance point Pc from point P0
a Distance from P0 to Pa
m
c
The mass of the platform without the driving
wheels and the rotors of the DC motors
m
w
The mass of each driving wheel plus the
rotor of its motor
I
C
The moment of inertia of the platform
without the driving wheels and the rotors of
the motors about a vertical axis through Pc
I
w
The moment of inertia of each wheel and the
motor rotor about the wheel axis
I
m
The moment of inertia of each wheel and the
motor rotor about a wheel diameter
Assuming that the velocity of
0
P
is in the direction
of x-axis of the local frame and there is no side slip,
and considering
[]
00
T
qxy
=
, the following
b
b
P
0
d
P
c
2
r
X
G
Y
G
y
0
x
0
y
x
Castor
Wheel
P
a
a
ϕ
ω
ICINCO2012-9thInternationalConferenceonInformaticsinControl,AutomationandRobotics
96
constraint with respect to
0
P
is obtained
00
sin cos 0xy
ϕϕ
−=
&&
(2)
By writing this constraint in matrix form, matrices
()
A
q and ()Sq are given by
[]
cos 0
() sin cos 0, () sin 0
01
Aq Sq
ϕ
ϕϕ ϕ
⎡⎤
⎢⎥
=− =
⎢⎥
⎢⎥
⎣⎦
(3)
Therefore, it can be written as
() () 0Aq Sq⋅= (4)
It is possible to write the kinematic equation of the
wheeled mobile robot motion in terms of the pseudo
velocities vector
()
nm
vt R
as
() ()
qSqvt=⋅
&
, (5)
where
[]
() () ()
T
vt vt t
ω
is made up of linear and
angular velocities. Taking the time derivative of (5)
() ()qSqvSqv=⋅+
&
&& &
(6)
Next, by replacing (5) and (6) into (1),
multiplying the result by
T
S
and considering (4), the
following equation can be obtained
() ()() ()
M
vt Cvvt Bq
τ
+=
&
, (7)
where
T
M
SMS=
,
()
T
CSMSCS=+
and
T
B
SB=
.
By denoting
()
B
q
τ
as
τ
() ()()Mv t C v v t
τ
+=
&
, (8)
The matrices
M
and
C
are obtained as follows:
=
0
0
,
̅
=
0

−
 0
(9)
where 2
cW
mm m=+ and
22
22
CmC W
I
IImdmb=+ + + .
There is a parametric vector
θ
on dynamics that
satisfies
() ()() (, ,,)
M
vt Cvvt Y qqvv
θ
+=
&&&
, (10)
where the parameters
,1,,4
i
i
θ
= K
are bounded and
defined as follows
123
, ,
c
mI md
θθθ
===
(11)
3 CONTROLLER DESIGN
3.1 Kinematic Controller
In the proposed control scheme, a kinematic
controller is used (Felipe et al., 2008). The design of
the kinematic controller is based on the kinematic
model of the WMR. The WMR’s kinematic model is
given by
cos sin
sin cos
01
xa
v
ya
ϕϕ
ϕϕ
ω
ϕ
⎡⎤
⎡⎤
⎢⎥
=
⎢⎥
⎢⎥
⎣⎦
⎢⎥
⎣⎦
&
&
&
, (12)
where
,
x
y
are the coordinates of the point of
interest
a
P
, and the outputs. By assuming
[]
,
T
hxy=
cos sin
sin cos
x
avv
hT
ya
ϕϕ
ϕϕ
ωω
⎤⎡
== =
⎥⎢
⎦⎣
&
&
&
, (13)
where
cos sin
sin cos
a
T
a
ϕϕ
ϕϕ
=
. (14)
The inverse of the matrix T is
1
cos sin
11
sin cos
T
aa
ϕϕ
ϕϕ
=
. (15)
Therefore, the inverse kinematics is given by
cos sin
11
sin cos
vx
y
aa
ϕϕ
ω
ϕϕ
⎡⎤
⎤⎡
⎢⎥
=
⎥⎢
⎢⎥
⎦⎣
⎢⎥
⎣⎦
&
&
, (16)
and the proposed kinematic controller is given by
tanh
cos sin
11
sin cos
tanh
x
dx
x
ref
ref
y
dx
y
k
x
Ix
I
v
k
yI y
aa
I
ϕϕ
ω
ϕϕ
+
=
+
⎡⎤
⎛⎞
⎜⎟
⎢⎥
⎡⎤
⎝⎠⎡⎤
⎢⎥
⎢⎥
⎢⎥
⎢⎥
⎢⎥
⎛⎞
⎣⎦
⎢⎥
⎣⎦
⎜⎟
⎜⎟
⎢⎥
⎝⎠
⎣⎦
&%
&%
(17)
Here,
d
x
xx=−
%
, and
d
yy y=−
%
are the current
position errors in the direction of
x
axis
and
yaxis
, respectively.
0
x
k >
and
0
y
k >
are the
gains of the controller,
x
I
R
, and
y
I
R
are
saturation constants, and
()
,
x
y
and
()
,
dd
x
y
are the
current and desired coordinates of the point of
interest, respectively. The purpose of the kinematic
TrajectoryTrackingControlofNonholonomicWheeledMobileRobots-CombinedDirectandIndirectAdaptiveControl
usingMultipleModelsApproach
97
Figure 2: Block diagram of the control architecture.
controller is to generate the reference linear and
angular velocities for the dynamic controller as
shown in Figure 2.
3.2 Adaptive Dynamic Controller
A Proportional-Integral (PI) filtered velocity
tracking error signal is given as (Wilson and
Robinett, 2001)
vv
s
eedt
λ
=+
(18)
where
λ
is a positive definite control gain and
velocity tracking error is defined as
vd
evv=−
. (19)
where
=
is the vector of the desired
linear and rotational velocities. Taking the derivative
of (18),
=
+e
(20)
can be obtained. Considering (8) and adding the PI
filtered error terms yields
2
() ( , )
dd
Ms C v s Y v v
θτ
+=
&&
(21)
2
(,) ( ) ()( )
dd d v d v
Yvv Mv e Cvv edt
θλ λ
=++ +
&&
(22)
To determine the control law and adaptive parameter
update rule, consider the following Lyapunov-like
function (Lewis et al., 2004)
1
1
2
TT
VsMs
θθ
=+Γ
%%
(23)
and differentiating the function with respect to time
1
1
2
TT T
VsMssMs
θθ
=++Γ
&
&
%%
&
&
(24)
By taking
M
s
&
from (21) and adding to (24), the
following equation can be obtained
()
2
1
(,) ()
1
2
T
dd
TT
VsYvv Cvs
sMs
θτ
θθ
=−
++Γ
&
&
&
&
%%
. (25)
By choosing the control law
2
ˆ
(,)
dd v
Yvv Ks
τθ
=+
&
(26)
and adding (26) into the (25), the following equation
can be obtained
()
2
1
(,)
1
2 ( )
2
TT
dd v
TT
VsYvv sKs
sM Cvs
θ
θθ
=−
+−+Γ
%
&
&
&
&
%%
(27)
Reader should note that the matrix
2()
M
Cv
&
is a
skew-symmetric matrix. By choosing the parameter
update rule as
()
12
(, ,) ( , )
TT
fdd
Yqvv Yvvs
θτ
=−Γ +
&
%
%
&
(28)
Kinematic
Controller
d
x
&
d
y
&
-
-
d
x
d
y
Dynamic
Controller
,
ref ref
v
ω
WMR
τ
Wheeled Mobile
Robot
, v
ω
Kinematics
1/s
1/s
1/s
x
&
y
&
ϕ
&
ϕ
x
y
ϕ
x
%
y
%
H(s)
Model 1
Model 2
M
Model N
±
±
±
1
I
e
2
I
e
N
I
e
f
τ
ˆ
j
θ
-
1/s
v
%
v
%
ϕ
ICINCO2012-9thInternationalConferenceonInformaticsinControl,AutomationandRobotics
98
with an identification error model
1
(, ,)
f
Yq vv
τθ
=
%
%
(29)
and inserting (28) into (27)
()
()
()
2
1
12
(,)
1
2 ( )
2
( , , ) ( , )
TT
dd v
T
TT T
fdd
VsYvv sKs
sM Cvs
Yqvv Yvvs
θ
θτ
=−
+−
+Γ−Γ +
%
&
&
&
%
%
&
(30)
where
1
(, ,)Yq vv
is the filtered regressor matrix
and
f
τ
is the filtered torque term (Ciliz and
Narendra, 1994). Rearranging (29)
()
1
1
2()
2
( , , )
TT
v
TT
VsKssMCvs
Yqvv
θτ
=− +
&
&
%
%
(31)
may be obtained. By considering the identification
error model in (29) and adding into (31)
()
11
1
2()
2
(, ,) (, ,)
TT
v
TT
VsKssMCvs
Y q vvY q vv
θθ
=− +
∫∫
&
&
%%
(32)
may be obtained. For the proof of stability, the same
procedures should be followed (Lewis et al., 2004).
It should be noted that
V
&
is negative definite. It can
be stated that
V
in (23) is upper bounded and that
()
M
q
is a positive definite matrix it can be stated
that
s
and
θ
%
are bounded. Standard linear control
arguments can be used to state that
v
e
and
v
e
are
bounded. Since
,,,
vv
ees
θ
%
are bounded it can be
shown that
s
&
and
V
&
are also bounded. The reader
should note that since
()
M
q
is lower bounded, it
can be stated that
V
is also lower bounded. Since
V
&
is lower bounded,
V
is negative definite and
V
&
is bounded, the Barbalat’s Lemma can be used to
state that
lim 0
t
V
→∞
=
&
(33)
which means that by Rayleigh-Ritz Theorem
{}
2
min
lim 0 or lim 0
v
tt
Ks s
λ
→∞ →∞
==
(34)
Using the standart linear control arguments the
following can be written
lim 0 and lim 0
v
v
tt
ee
→∞ →∞
==
(35)
3.3 Adaptive Dynamic Controller with
Multiple Models
Identification models have the following structure
ˆ
ˆ
ˆ
ˆ
() ()() (, ,,)
jj j j
M
vt C vvt Y qqvv
τθ
=+ =
&&&
(36)
where
1, ,
j
N= K
,
ˆ
j
θ
denoting the parameter
estimate vector and
(,,,)Yqqvv
&&
is the non-linear
regressor matrix. The regressor matrix common to
all models, but the parameter vector
ˆ
j
θ
has different
initializations chosen from a given compact
parameter set. Using the filtering technique
previously mentioned nonlinear regressor matrix
without acceleration signal can be obtained and will
be denoted as
1
(, ,)Yq vv
. Each model is updated
using simple gradient algorithm as it is in single
model case:
12
((,,) (,))
j
TT
jfdd
Yqvv Yvvs
θτ
=−Γ +
&
%
%
&
(37)
based on the error model which is defined as,
1
ˆ
(, ,)
jj j
f
Iff j
eYqvv
τττ θ
==−=
%
%
(38)
where
f
τ
%
is the filtered torque prediction error.
2
(,)
rr
Yvv
&
is the regressor matrix common to all
models which is given in (22). The torque vector
j
τ
of
jth
identification model is given as:
2
ˆ
(,)
jddjv
Yv v Ks
τθ
=+
&
. (39)
Adding the equations and (21) into (8), the closed
loop dynamics can be obtained as:
() ( ) ()( )
jj v jd jdv
v
M
sCvsKs Mv e Cvv e
λλ
++=++ +
%
%
&&
(40)
which can further be written as
2
() ( , )
jj v ddj
Ms C vs Ks Y v v
θ
++=
%
&&
(41)
At any instant, the identification errors of the
N
models are available, but only one of the torque
vectors
j
τ
is chosen as the input to the WMR.
In order to choose a switching criterion, first a
permissible switching sequence and a switching rule
must be given (Ciliz and Narendra, 1994 & 1995). A
finite or infinite sequence
+
RTT
ii
:
is defined as a
switching sequence if
0
0T =
and
1
,
ii
iT T
+
∀<
.
Additionally, if there is a number
min
0T >
such that
1min
,
ii
iT T T
+
∀−
, then the sequence is called
permissible switching scheme.
A switching rule is a function of time that takes
values in the set
1, ,
N
K
is constant in
[
)
1
,
ii
TT
+
and
is continuous from right. In other words, a function
(): 1, ,ht R N
+
K
is called switching rule, if there
exists a switching sequence
0i
i
T
=
such that if
[
)
1
,
ii
tTT
+
for some
i <∞
, then
() ( )
i
ht hT=
. With
this definition torque input in (21) can be defined as:
TrajectoryTrackingControlofNonholonomicWheeledMobileRobots-CombinedDirectandIndirectAdaptiveControl
usingMultipleModelsApproach
99
()
( ) ( ) 0.
ht
ttt
ττ
=≥
(42)
The torque vector combined with a permissible
switching rule given as
()
() 2
ˆ
(,)
ht
ht d d j v
Yvv Ks
τθ
=+
&
(43)
For the proof of stability, the same procedure will be
followed as in the single model case. The additional
requirement is that under any permissible switching
rule, all signals should remain bounded. We have a
Lyapunov-like function
1
1
2
TT
jjjj
VsMs
θθ
=+Γ
%%
(44)
The derivative of (44) can be obtained as in the
following equation
()
11
1
2()
2
(, ,) (, ,)
TT
jv jj
TT
jj
VsKssMCvs
YqvvYqvv
θθ
=− +
∫∫
&
&
%%
(45)
j
V
&
is negative definite. It can be stated that
j
V
in
(44) is upper bounded and that
()
j
M
q
is a positive
definite matrix, it can be stated that
s
and
j
θ
%
are
bounded. Standard linear control arguments can be
used to state that
v
e
and
v
e
are bounded. Since
,,,
vv
ees
θ
%
are bounded it can be shown that
s
&
and
j
V
&
are also bounded. The reader should note that
since
()
j
M
q
is lower bounded, it can be stated that
j
V
is also lower bounded. Since
j
V
&
is lower
bounded,
j
V
is negative definite and
j
V
&
is bounded,
the Barbalat’s Lemma can be used to state that
lim 0
j
t
V
→∞
=
&
, (46)
which means that by Rayleigh-Ritz Theorem
{}
2
min
lim 0 or lim 0
v
tt
Ks s
λ
→∞ →∞
==. (47)
Using the standard linear control arguments as in
single model case the following can be written
lim 0 and lim 0
v
v
tt
ee
→∞ →∞
==
. (48)
3.4 Proof of Stability for the Kinematic
Controller
In order to understand the rest of the proof, the
reader may read (Martins et al., 2008). By
considering (13) and (14):
1
2
tanh
0
0
tanh
x
x
x
y
y
y
k
x
I
I
x
I
y
k
y
I
ε
ε
⎡⎤
⎛⎞
⎢⎥
⎜⎟
⎡⎤
⎡⎤
⎝⎠
⎡⎤
⎢⎥
+=
⎢⎥
⎢⎥
⎢⎥
⎢⎥
⎛⎞
⎣⎦
⎣⎦
⎣⎦
⎢⎥
⎜⎟
⎜⎟
⎢⎥
⎝⎠
⎣⎦
%
&
%
&
%
%
(49)
One can see that the error vector
ε
can also be
written as
Te
, where
e
is the velocity tracking error
and matrix
T is defined before. Rewriting (49)
() ,
tanh
0
()
0
tanh
x
x
x
y
y
y
hLh Te
k
x
I
I
Lh
I
k
y
I
+=
⎛⎞
⎜⎟
⎡⎤
⎝⎠
=
⎢⎥
⎛⎞
⎣⎦
⎜⎟
⎜⎟
⎝⎠
&
%%
%
%
%
(50)
Now considering Lyapunov candidate function
and its derivative
()
1
,
2
()
T
TT
Vhh
VhhhTeLh
=
==
%%
&
%% % %
&
(51)
and a sufficient condition for
0V <
&
can be
expressed as
()
TT
hLh hTe>
%% %
(52)
For small values of the control error h
%
following
can be written
0
() ,
0
x
xy xy
y
k
Lh K h K
k
⎡⎤
≈=
⎢⎥
⎣⎦
%%
K
(53)
Now the sufficient condition for
0V <
&
can be
written as
2
,
min( , ) ,
min( , )
TT
xy
xy
xy
hK h hTe
kk h hTe
Te
h
kk
>
>
>
%%%
%%
%
(54)
It is shown that
e
tend to zero as t →∞, which
implies that condition in (43) is verified for any
value of h
%
. Thus,
() 0ht
%
as t →∞.
3.5 Switching Criterion
A cost function is considered in the form
()
12
0
() () () () ()
jj j j
t
TtT
jI I I I
J
t etGet e etGetd
λτ
τ
−−
=+
(55)
where
j
J
is the cost function of the j
th
model,
j
I
e
is
ICINCO2012-9thInternationalConferenceonInformaticsinControl,AutomationandRobotics
100
the identification error associated with the j
th
model,
12
,
nn
GG R
×
are positive (semi)-definite weight
matrices and
0
λ
is a scalar forgetting factor.
a
J
is denoted as the cost function of the current model.
If
() ()
aj
Jt Jt>
with defined switching sequence, it
means that adaptive model must be switched to the
j
th
model according to the switching criterion.
4 SIMULATIONS
In the simulations, the WMR should track an eight-
shaped trajectory given by
()
()
sin 2 ,
sin
rg r
rg r
x
xR t
yyR t
ω
ω
=+
=+
(56)
where
2.5
g
x =
,
5.5
g
y =
,
0.04
r
ω
=
and
7.5R =
.
Initially robot
0
2x =
and
0
6.5y =
and robot has
zero velocities and
6
π
ϕ
=−
.
The parameters of the WMR are taken as
2
0.0025 .
m
I
Kg m=
,
2
15.625 .
c
I
Kg m=
,
0.15rm=
,
0.75bm=
,
0.3 am=
,
0.3dm=
,
0.1
L
m=
,
1
w
mKg=
,
36
c
mKg=
,
2
0.005 .
w
I
Kg m=
,
10
v
K =
,
(10,10)diag
λ
=
,
(2,2,2)diagΓ=
,
1
α
=
,
10
x
k =
,
10
y
k =
,
1
y
I =
,
1
x
I =
. The switching
sequence has a time step of 5 ms.
The real values of the unknown parameters are
[38 19.95 10.8]
T
θ
= , and the initial estimates for
the parameters are
ˆ
[20 7 3]
T
θ
=
.
In order to show effectiveness of the developed
solution ten identification model has been chosen as
=
29 11 5
,
=
32 14 7
,
=
35 17 9
,
=
38 20 11
,
=
41 23 13
,
=
44 26 15
,
=
47 29 17
,
=
50 32 19
,
=
53 35 21
,

=
56 38 23
It can be seen from the figures that proposed control
approach enhances the performance of both velocity
tracking and trajectory tracking. In Fig. 3, there is a
trajectory tracking results for both single model and
multiple model cases. The controller provides the
reference trajectory tracking with a similar
performance for two cases. However, if one focus on
the trajectories for the first five seconds as seen in
Fig. 4, he can see the differences. Also, Fig. 5
shows the tracking errors on the x and y axis. In Fig.
6 and 8, there are linear and rotational velocity
errors, respectively. In order to show the
enhancement of the transient behaviour, Fig. 7 and 9
shows the linear and rotational errors for the first 5
seconds of the simulation. Similarly, Fig. 10-13
show the results for the integral of the linear and
rotational velocities. Fig. 14 shows the switching
between models during the simulation.
Figure 3: Robot position in single model case and multiple
model case vs. Reference Trajectory.
Figure 4: Robot position in single model case and multiple
model case vs. reference trajectory (five seconds to see the
effect).
TrajectoryTrackingControlofNonholonomicWheeledMobileRobots-CombinedDirectandIndirectAdaptiveControl
usingMultipleModelsApproach
101
Figure 5: Position Errors on the x and y axis.
Figure 6: Linear velocity tracking error in single model
case and multiple models case.
Figure 7: Linear velocity tracking error in single model
case and multiple models case (five seconds to see the
effect).
Figure 8: Rotational velocity tracking error in single
model case and multiple models case.
Figure 9: Rotational velocity tracking error in single
model case and multiple models case (five seconds to see
the effect).
Figure 10: Integral of linear velocity tracking error in
single model case and multiple models case.
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
Tim e
Position errors on the x and y axis
Position Error on the X-Axis (Single Model)
Position Error on the Y-Axis (Single Model)
Position Error on the X-Axis (Multiple Models)
Position Error on the Y-Axis (Multiple Models)
0 20 40 60 80 100 120 140 160
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
Time
Tracking error1
Linear Velocity Tracking Error (Single Model)
Linear Velocity Tracking Error (Multiple Models)
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
Tim e
Tracking error1
Linear Velocity Tracking Error (Single Model)
Linear Velocity Tracking Error (Multiple Models)
0 20 40 60 80 100 120 140 160
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
Tim e
Tracking error2
Rotational Velocity Tracking Error (Single Model)
Rotational Velocity Tracking Error (Multiple Models)
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
Tim e
Tracking error2
Rotational Velocity Tracking Error (Single Model)
Rotational Velocity Tracking Error (Multiple Models)
0 20 40 60 80 100 120 140 160
-0.05
0
0.05
0.1
0.15
0.2
0.25
0.3
Tim e
Integral of the Tracking error1
Integral of Linear Velocity Tracking Error (Single Model)
Integral of Linear Velocity Tracking Error (Multiple Models)
ICINCO2012-9thInternationalConferenceonInformaticsinControl,AutomationandRobotics
102
Figure 11: Integral of linear velocity tracking error in
single model case and multiple models case (ten seconds
to see the effect).
Figure 12: Integral of rotational velocity in single model
case and multiple models case.
Figure 13: Integral of rotational velocity in single model
case and multiple models case (ten seconds to see the
effect).
Figure 14: Switching between models.
5 CONCLUSIONS
An adaptive control algorithm with a multiple
models approach is proposed for the trajectory
tracking of a WMR. The controller uses a combined
direct and indirect adaptive control approach where
both prediction and tracking errors are used in
identification and switches between multiple models
of the WMR dynamics and the control input is
applied based on the model which closely describes
the WMR dynamics. This dynamic controller
provides fast velocity tracking under parameter
uncertainties. The proposed kinematic controller
provides the velocity profile needed for the
trajectory tracking of the WMR in Cartesian
coordinates. The stability of the overall control
system was proved. As a result, simulations show
that the proposed control system is applicable to the
WMR and it significantly enhances the transient
behavior during the trajectory tracking.
REFERENCES
Cezayirli, A., Ciliz, K., (2004). Multiple Model Based
Adaptive Control of a DC Motor Under Load
Changes., Proceedings of the IEEE International
Conference on Mechatronics.
Cezayirli, A., Ciliz, M. K., (2007). Transient Performance
Enhancement of Direct Adaptive Control of Nonlinear
Systems Using Multiple Models and Switching. IET
Control Theory Appl., 1(6), 1711-1725.
Cezayirli, A., Ciliz, M. K., (2008), Indirect Adaptive
Control of Non-Linear Systems Using Multiple
Identfication Models and Switching. International
Journal of Control, 81(9), 1434-1450.
Chen, W., Anderson, B. D. O., (2009). Multiple Model
Adaptive Control (MMAC) for Nonlinear Systems
with Nonlinear Parametrization. Joint 48
th
IEEE
0 1 2 3 4 5 6 7 8 9 10
0
0.05
0.1
0.15
0.2
0.25
Time
Integral of the Tracking error1
Integral of Linear Velocity Tracking Error (Single Model)
Integral of Linear Velocity Tracking Error (Multiple Models)
0 20 40 60 80 100 120 140 160
-0.08
-0.06
-0.04
-0.02
0
0.02
0.04
0.06
0.08
0.1
Time
Integral of the Tracking error2
Integral of the Rotational Velocity Tracking Error (Single Model)
Integral of the Rotational Velocity Tracking Error (Multiple Models)
0 1 2 3 4 5 6 7 8 9 10
-0.08
-0.06
-0.04
-0.02
0
0.02
0.04
0.06
0.08
0.1
Time
Integral of the Tracking error2
Integral of Rotational Velocity Tracking Error (Single Model)
Integral of Rotational Velocity Tracking Error (Multiple Models)
TrajectoryTrackingControlofNonholonomicWheeledMobileRobots-CombinedDirectandIndirectAdaptiveControl
usingMultipleModelsApproach
103
Conference on Decision and Control and 28
th
Chinese
Control Conference, Shangai, P.R. China.
Ciliz, K., Narendra, K. S., (1994). Multiple Model Based
Adaptive Control of Robotic Manipulators.
Proceedings of the 33
rd
Conference on Decision and
Control. Lake Buena Vista, FL.
Ciliz, K., Narendra, K. S., (1995). Intelligent Control of
Robotic Manipulators: A Multiple Model Based
Approach. Intelligent Robots and Systems 95. 'Human
Robot Interaction and Cooperative Robots',
Proceedings.
Ciliz, K., Cezayirli, A., (2004). Combined Direct and
Indirect Control of Robot Manipulators Using
Multiple Models. Proceedings of the 2004 IEEE
Conference on Robotics, Automation and
Mechatronics, Singapore.
Ciliz, K., Tuncay, M. Ö., (2005). Comparative
Experiments with a Multiple Model Based Adaptive
Controller for SCARA Type Direct Drive
Manipulator. Robotica, 23(6), 721-729.
Ciliz, M. K., Cezayirli, A., (2006). Increased Transient
Performance for Adaptive Control of Feedback
Linearizable Systems Using Multiple Models.
International Journal of Control, 79(10), 1205-1215.
Ciliz, K., Cezayirli, A., (2006). Adaptive Tracking for
Nonlinear Plants Using Multiple Identification Models
and State-Feedback. IEEE Industrial Electronics,
IECON 2006 - 32nd Annual Conference.
De La Cruz, C., Carelli, R., (2006). Dynamic Modeling
and Centralized Formation Control of Mobile Robots.
IEEE Industrial Electronics, IECON 2006.
D’Amico, A., Ippoliti, G., Longhi, S., (2006). A Multiple
Models Approach for Adaptation and Learning in
Mobile Robots Control. Journal of Intelligent Robot
Systems, 47, 3-31.
De La Cruz, C., Carelli, R., Bastos, T. F., (2008).
Switching Adaptive Control of Mobile Robots.
Industrial Electronics, 2008, ISIE 2008, IEEE
International Symposium.
Fierro, R., Lewis, F. L., (1995). Control of a Nonholonomic
Mobile Robot: Backstepping Kinematics into
Dynamics. Proceedings of the 34
th
Conference on
Decision and Control. New Orleans, LA.
Fukao, T., Nakagawa, H., Adachi, N., (2000). Adaptive
Tracking Control of a Nonholonomic Mobile Robot.
IEEE Transactions on Robotics and Automation,
16(5), 609-615.
Gholipour, A., Yazdanpanah, M. J., (2003). Dynamic
Tracking Control of Nonholonomic Mobile Robot
with Model Reference Adaptation for Uncertain
Parameters. European Control Conference
Cambridge, UK.
Islam, S., Liu, P. X., (2009). Adaptive Output Feedback
Control for Robot Manipulators Using Lyapunov-
Based Switching. The 2009 IEEE/RJS International
Conference on Intelligent Robots and Systems, St.
Louis, USA.
Kalkkuhl, J., Johansen, T. A., Lüdemann, J., (2002).
Improved Transient Performance of Nonlinear
Adaptive Backstepping Using Estimator Resetting
Based on Multiple Models. IEEE Transactions on
Automatic Control, (47)1, 136-140.
Kanayama, Y., Kimura, Y., Miyazaki, F., Noguchi, T.,
(1990). A Stable Tracking Control Method for an
Autonomous Mobile Robot. Robotics and Automation
International Conference. USA.
Lee, C. Y., (2006). Adaptive Control of A Class of
Nonlinear Systems Using Multiple Parameter Models.
International Journal of Control, Automation, and
Systems, 4(4), 428-437.
Martins, F. N., Celeste, W. C., Carelli, R., Sarcinelli-Filho,
M., Bastos-Filho, T., (2008), An Adaptive Dynamic
Controller for Autonomous Mobile Robot Trajectory
Tracking. Control Engineering Practice, 16, 1354-1363
Narendra, K. S., Balakrishnan, J., (1997). Adaptive
Control Using Multiple Models. IEEE Transactions on
Automatic Control, 42(2), 171-187.
Narendra, K. S., George, K., (2002). Adaptive Control of
Simple Nonlinear Systems Using Multiple Models.
Proceedings of the American Control Conference
Anchorage.
Park, B. S., Park, J. B., Choi, Y. H., (2011). Adaptive
Observer-Based Trajectory Tracking Control of
Nonholonomic Mobile Robots. International Journal
of Control, Automation, and Systems, 9(3), 534-541.
Petrov, P., (2010), Modeling and Adaptive Path Control of
a Differential Drive Mobile Robot. Proceedings of the
12th WSEAS International Conference on Automatic
Control, Modellig and Simulation.
Pourboghrat, F., Karlsson, M. P., (2002). Adaptive
Control Of Dynamic Mobile Robots with
Nonholonomic Constraints. Computers & Electrical
Engineering, 28(4), 241-253.
Shojaei, K., Shahri, A. M., Tarakameh, A., Tabibian, B.,
(2011). Adaptive Trajectory Tracking Control of a
Differential Drive Wheeled Mobile Robot. Robotica,
29(3), 391-402.
Wison, D. G., Robinett, R. D., (2001). Robust Adaptive
Backstepping Control for a Nonholonomic Mobile
Robot. IEEE International Conference on Systems,
Man, and Cybernetics.
Ye, X., (2008), Nonlinear Adaptive Control Using
Multiple Identification Models. Systems & Control
Letters, 3, 488-491.
Yun, X., Yamamoto, Y., (1992). On Feedback
Linearization of Mobile Robots. Department of
Computer & Information Science Technical Reports.
University of Pennsylvania.
Zhengcai, C., Yingtao, Z., Qidi, W., Adaptive Trajectory
Tracking Control for a Nonholonomic Mobile Robot.
Chinese Journal of Mechanical Engineering, 24(3), 1-7.
ICINCO2012-9thInternationalConferenceonInformaticsinControl,AutomationandRobotics
104