This article has been accepted for inclusion in a future issue of this journal.

Content is final as presented, with the exception of pagination.

IEEE/ASME TRANSACTIONS ON MECHATRONICS 1
Robust Adaptive Controller for a TractorTrailer
Mobile Robot
Ali Keymasi Khalaji and S. Ali A. Moosavian
AbstractA tractortrailer wheeled robot (TTWR) is a kind
of modular robotic system that consists of a tractor template at-
tached with a single or multiple trailers, and hence it is a nonlinear
and underactuated system subjected to nonholonomic constraints.
Tracking control of such a complicated system is a challenging
problem, and that is the focus of this paper. To this end, rst dy-
namics model of a TTWR is developed. Next, feasible reference
trajectories are generated to dene a trajectory tracking problem.
Then, a Lyapunov kinematic control law is elaborated to stabi-
lize tracking errors. Subsequently, a feedback linearizing dynamic
controller (FLDC) is designed to generate actuator torques. In a
wheeled mobile robot (WMR), like most of real engineering appli-
cations, it is impossible to obtain an exact dynamics model due to
various unknown, or unpredictable and irregular features. There-
fore, the robustness of controllers to overcome uncertainties and
external disturbances is necessary. So, a robust adaptive feedback
linearizing dynamic controller (RAFLDC) is proposed to control
the system using estimated upper-bounds of uncertainties. The
stability of the control algorithm is veried using the Lyapunov
method. Robustness and effectiveness of the proposed controller,
and comparison of results for RAFLDC and FLDC algorithms, is
investigated using both simulation studies and experimental imple-
mentations, and obtained results will be discussed.
Index TermsAdaptive control, nonholonomic system, robust
control, trajectory tracking, wheeled mobile robot (WMR).
I. INTRODUCTION
M
OBILE robots have attracted much attention in industry
and research in recent decades, [1]. Various locomo-
tion systems have been proposed for mobile robots in different
environments, [2][4]. The wheel is the most popular locomo-
tion mechanism in mobile robotics and man-made vehicles due
to its simplicity, efciency, and exibility. A wheeled mobile
robot (WMR) as a result of nonslip and pure rolling condi-
tions of wheels is subjected to nonholonomic constraints. In [5],
classication of kinematics and dynamics models for various
types of WMRs has been analyzed. The control of nonholo-
nomic systems is a challenging problem as a result of system
inherent characteristics, [6]. Highly nonlinear dynamic models,
nonsquare multiinput multioutput models, underactuated, and
driftless mechanical systems are some of their attributes. Possi-
ble motion tasks can be classied into point stabilization, path
following, and trajectory tracking. Trajectory tracking is one of
Manuscript received December 9, 2012; revised March 12, 2013; accepted
April 23, 2013. Recommended by Technical Editor G. Herrmann.
The authors are with K. N. Toosi University of Technology, Tehran 19697
64499, Iran (e-mail: [email protected]; [email protected]).
Color versions of one or more of the gures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identier 10.1109/TMECH.2013.2261534
the major control problems in this eld that has attracted much
attention, [7]. In this problem, the robot must follow a desired
Cartesian path with a specied timing law, [8]. However, most
of the controllers are proposed for autonomous navigation of
unicycle type or carlike mobile platforms, while there exist few
reports on mobile robots towing passive trailers.
The transportation capacity can be increased using trailers,
while the cost of the car with trailers is much lower than the
cost of multiple cars. The aims of control research for tractor
trailer systems vary from motion aid in human-driven trans-
portation and delivery systems, to fully autonomous navigation
in multibody mobile robotics. However, the major drawback
of the trailer systems is that the control problem is difcult.
In [9], some properties and kinematic aspects of such systems,
in [10] mechanical design of these systems, in [11] steering
limitations, in [12] the controllability of these systems using
differential geometric tools have been reviewed and discussed.
Nonholonomic motion planning and the concept of atness for
multibody WMRs have been discussed in [13] and [14]. In [15],
the motion-planning problem for a car with multiple trailers is
investigated. Majorities of the works in this eld are on motion
planning problemand the concept of atness and there exist few
works on control of these systems, which is the focus of this
paper.
In [16], virtual steering is described for backward motion
control of a tractortrailer system. On the other hand [17] uses
model-based fuzzy-control for backing-up control of a vehicle
with triple trailers. In [18] and [19], kinematic controllers have
been proposed for tracking control of off-axle articulated vehi-
cles. A kinematic control law is proposed for path tracking of
articulated vehicles using exact linearization in [20]. In [21], a
kinematic controller is proposed for path tracking of a tractor
trailer system along the rectilinear and circular paths. Also, a
kinematic controller for trajectory tracking problem was pre-
sented in [22] based on the time-varying linear quadratic regu-
lator method. Besides the works on automated wheeled robots,
there exist some research works in the vehicle community for
truck and trailer systems, e.g. [23]. The proposed controllers are
mainly on kinematic models of multibody WMRs, assuming
perfect velocity tracking and neglecting inertia effects. How-
ever, in real applications particularly in high speed motions
and higher inertias, it is necessary to exploit a dynamic control
approach as discussed in [24]. However, there exist very few
control approaches for dynamic control of these systems, which
is elaborated in is paper.
Robustness to uncertainties and insensitivity to disturbances
for control algorithms is essential in actual applications. Prior
knowledge of upper-bounds of uncertainties is often needed to
1083-4435/$31.00 2013 IEEE
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
2 IEEE/ASME TRANSACTIONS ON MECHATRONICS
apply robust controllers for robotic systems, though such bounds
cannot be easily obtained. On the other hand, most of the ro-
bust controllers are conservative because prior knowledge of
bounds of uncertainties in worst case scenario is required, [25].
Therefore, combined robust adaptive controllers were proposed
to have the benets and overcome the drawbacks of these meth-
ods, [26]. In [27] and [28], some robust adaptive controllers
have been proposed for nonlinear systems in the presence of
parametric uncertainties. In [29], some concepts and methods
of adaptive and robust control algorithms have been reviewed
and discussed for nonlinear systems. As a matter of fact adaptive
algorithms can be used to estimate upper-bounds of uncertain-
ties [30]. Also, in [31] and [32] authors propose soft-computing
methods to approximate uncertain functions. In [33], an adap-
tive terminal sliding mode tracking control design for a class
of nonlinear systems is presented using fuzzy logic. In [34],
an adaptive sliding mode controller approach is presented for
tracking control of robotic manipulators.
As discussed previously, most of the works on tractortrailer
systems are on motion-planning problemand the concept of at-
ness, and hardly on kinematic control of these systems, while
in high speed motions and higher inertias a dynamic control
approach should be employed. Therefore, this paper is one of
the rsts to develop a robust adaptive dynamic control for non-
holonomic tractortrailer wheeled mobile robots (TTWRs) with
experimental implementation. In fact, the main achievements of
this paper which will be presented in the following sections are:
1) TTWR dynamics model is developed;
2) a kinematic controller is designed to stabilize tracking er-
rors, based on the feasible reference trajectories generated
for the TTWR;
3) upper-bounds for the system uncertainties are analytically
determined;
4) a robust adaptive feedback linearizing dynamic controller
(RAFLDC) is presented in order to control the TTWR at
a dynamic level in the presence of external disturbances
and parameter uncertainties;
5) appropriate adaptive rules are designed to compensate the
TTWR upper-bounded lumped uncertainties;
6) stability of the proposed control algorithm is proved using
the Lyapunov method;
7) both simulation and experimental studies are presented,
and the comparative results are discussed.
II. SYSTEM DESCRIPTION AND KINEMATIC MODELING
A layout of the considered TTWR system is displayed in
Fig. 1, which includes a differentially driven vehicle towing a
passive trailer. The tractor is equipped with two driving wheels
while a spherical wheel is used for stable motion, and the trailer
has two passive wheels. The tractor and trailer are connected to
each other via a revolute joint at point Qas shown in Fig. 2. Nor-
mally, it is assumed that d
0
= 0, except in the case of robustness
to un-modeled dynamics in Section VII, where we have stud-
ied off-axle hitching problem. Here, it is assumed that wheels
without slip in the lateral direction are considered as a thin solid
disk having a single point contact with motion surface; while
Fig. 1. Experimental setup.
Fig. 2. Differentially driven wheeled mobile robot towing a trailer.
the motion is planar. A coordinate system (X, Y ) is considered
as the inertial frame. Points C
0
and C
1
represent the tractor and
trailer centroids, respectively, also
r
and
1
represent angular
displacements of tractors right and left wheels, respectively. d
denotes the distance between points P
0
and P
1
in d
0
= 0 case,
and also a
0
and a
1
are the distances between points P
0
and C
0
and points P
1
and C
1
, respectively, as shown in Fig. 2.
System constraints can be written in matrix form as follows:
A(q) q=
_
sin
0
cos
0
d cos (
0

1
) 0
sin
1
cos
1
0 0
_
q = 0
(1)
where q = (x, y,
1
,
0
)
T
is system conguration vector. (x, y)
is the coordinate of point P
1
in the inertial frame,
0
and
1
represent the orientation of the tractor and trailer with respect to
the inertial frame, respectively. Also, A(q) is system constraint
matrix.
Kinematic equations of the mobile robot can be written as
q(t) = S(q)u (2)
where u = (u
1
u
2
)
T
describes an independent set of variables
which is here the system input vector. u
1
is the linear velocity of
point P
1
and u
2
is angular velocity of the tractor. These inputs
are related to angular velocities of differentially actuated wheels
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
KEYMASI KHALAJI AND MOOSAVIAN: ROBUST ADAPTIVE CONTROLLER FOR A TRACTORTRAILER MOBILE ROBOT 3
as follows:
_
_
_
u
1
=
r
2
(
r
+
l
) cos(
0

1
)
u
2
=
r
2b
(
r
+
l
)
(3)
where r is the radius of wheels, b is the half of the distance
between tractor actuated wheels, and
r
and
1
denote angular
displacements of tractors right and left wheels, respectively.
Also, S(q) matrix can be found as
S(q) = [ S
1
(q) S
2
(q) ]
=
_
cos
1
sin
1
1
d
tan (
0

1
) 0
0 0 0 1
_T
. (4)
Moreover, we can show the following, which describes the
natural orthogonal complement property [2]:
S
T
(q) A
T
(q) = 0. (5)
III. KINEMATIC CONTROLLER DESIGN
In this section, rst reference trajectory is planned for the
TTWR. Next, a stable kinematic control law is designed for the
system.
A. Trajectory Planning
To obtain admissible trajectories for the trajectory tracking
problem of the TTWR, it should be noted that an admissible
trajectory is a solution of the differential system corresponding
to the kinematic model of the mobile robot (including the con-
straints), [13]. Assuming the Cartesian coordinates of the point
P
1
as at outputs, the TTWR will be a at system [14]. A useful
tool that allows testing the controllability of driftless nonlinear
systems is the Lie Algebra rank condition, [35]. Flatness implies
that, for generic states, the rank of the Lie algebra generated by
the two vector elds associated to the control variables is full and
the system is controllable, [15]. Flat outputs can be expressed
as follows:
x = x(t), y = y(t). (6)
Consequently, the system conguration vector q and system
kinematic input vector u can be expressed algebraically as a
function of at outputs and their derivatives up to certain order
[36].
Therefore, from the rst two equations of (2), it can be con-
cluded that

1
= ATAN2{ y, x}. (7)
Using the third equation of (2), we have

0
=
1
+ atan
_
d

1
u
1
_
. (8)
Also, the rst two equations of (2) yield
u
1
=
_
x
2
+ y
2
. (9)
Substitution from (7) into (8) and differentiation yields the
second input as
u
2
=

1
+ d u
1
( y x x y) u
2
1
3( y x x y) ( x x + y y)
u
6
1
+ d
2
( y x x y)
2
u.
(10)
The robot reference outputs in Cartesian coordinates are ex-
pressed as follows:
x
r
= x
r
(t), y
r
= y
r
(t) (11)
where the index r is specied to system variables on
reference trajectories. Therefore, the reference trajectories

0r
(t),
1r
(t), u
1r
(t) and u
2r
(t) can be calculated from (7) to
(10).
Property 1: Reference trajectories x
r
(t), y
r
(t),
0r
(t) and

1r
(t) and reference velocity inputs and their derivatives are
continuous and uniformly bounded and reference velocity in-
puts do not tend to zero as t .
B. Kinematic Control Law
In this section, a control law is designed for trajectory
tracking of the mobile robot. First, an error dynamics is de-
veloped for tracking problem. Then, a feedback control law
u = u(q, q, q
r
, u
r
) is designed to stabilize tracking errors
e = q q
r
to the origin.
First an error vector e is dened in a new space as
e = T e (12)
where transformation matrix T and tracking error array e are as
T =
_
_
_
_
_
cos
1r
sin
1r
0 0
sin
1r
cos
1r
0 0
0 0 1 0
0 0 0 1
_
_
_
_
_
; e =
_
_
_
_
_
e
x
e
y
e

0
e
1
_
_
_
_
_
. (13)
Differentiation from (12) yields
_

_
e
x
= u
2r
e
y
+ u
1
cos e

1
u
1r
e
y
= u
2r
e
x
+ u
1
sin e

1
e

0
= u
2
u
2r
e

1
=
u
1
d
tan(
0

1
)
u
1r
d
tan(
0r

1r
).
(14)
In order to determine the control inputs u
1
and u
2
, we suppose
the following diffeomorphism:
_

_
z
1
= e
x
z
2
= e
y
z
3
= tan e

1
z
4
=
tan(
0

1
) cos e

1
tan(
0r

1r
)
d cos
3
e

1
+ e
y
(15)
with the following input transformations:
_
w
1
= u
1
cos e

1
u
1r
w
2
= dz
4
/dt.
(16)
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
4 IEEE/ASME TRANSACTIONS ON MECHATRONICS
With these change of variables the tracking error dynamics
will be transformed to the following system:
_

_
z
1
= u
2r
z
2
+ w
1
z
2
= u
2r
z
1
+ (u
1r
+ w
1
)z
3
z
3
= u
1r
(z
4
z
2
) + w
1
_
z
4
z
2
+
tan(
0r

1r
)
d
_
1 + z
2
3
_
_
z
4
= w
2.
(17)
Now we dene the following control inputs
_
_
_
w
1
= k
1
|u
1r
|
_
z
1
+ z
3
_
z
4
+
_
1 + z
2
3
_
tan (
0r

1r
)
d
__
w
2
= k
2
u
1r
z
3
k
3
|u
1r
| z
4
.
(18)
Proposition 1: The control law(18) guarantees the asymptotic
stabilization of the dynamic system described by (17) to the
origin as t .
Proof: Considering the following positive denite function
V
1
=
1
2
_
z
2
1
+ z
2
2
+ z
2
3
+
1
k
2
z
2
4
_
(19)
the time derivative of V
1
can be found as

V
1
=k
1
|u
1r
|
_
z
1
+ z
3
_
z
4
+
_
1 + z
2
3
_
tan (
0r

1r
)
d
__
2

k
3
k
2
|u
1r
| z
2
4
. (20)
It is assumed that k
1
, k
2
, and k
3
are positive constants. There-
fore, it can be concluded that

V
1
is a negative semidenite
function. Therefore, V
1
is a nonincreasing function. Hence,
z
i
(i : 1 4) are globally bounded. Using Property 1, u
1r
is
continuous, and its derivative is globally bounded. Therefore,
|u
1r
| is globally Lipschitz and thus uniformly continuous, [37].
On the other hand, since z
i
(i : 1 4) are globally bounded, as-
suming |
0r

1r
| <

2
it can be concluded that z
i
(i : 1 4)
are globally bounded. Therefore, z
i
(i : 1 4) are globally Lip-
schitz and consequently uniformly continuous, [37]. Therefore,

V
1
is uniformly continuous and by application of Barbalats
Lemma,

V
1
tends to zero when t tends to innity, [25]. Thus, us-
ing (20) and using the assumption that u
1r
does not tend to zero,
we can conclude that z
4
and z
1
+ z
3
(z
4
+ (1 + z
2
3
)
tan(
0 r

1 r
)
d
)
(and thus z
1
+ z
3
(1 + z
2
3
)
tan(
0 r

1 r
)
d
) tend to zero. On the
other hand, using (17) and (18) we can write
d
dt
_
u
2
1r
z
4
_
= 2u
1r
u
1r
z
4
u
2
1r
(k
2
u
1r
z
3
+ k
3
|u
1r
| z
4
) .
(21)
Since z
4
tends to zero, we can conclude
lim
t
_
d
dt
_
u
2
1r
z
4
_
+ k
2
u
3
1r
z
3
_
= 0. (22)
Since u
2
1r
z
4
is uniformly continuous and converges to zero,
using Barbalats Lemma it can be shown that
d
dt
_
u
2
1r
z
4
_
also
converges to zero. Thus, using (22) and the assumption that u
1r
does not tend to zero from property 1, one obtains that z
3
(and
thus z
1
also) tends to zero.
On the other hand, using (17) and (18), we can write
d
dt
_
u
2
1r
z
1
_
= 2u
1r
u
1r
z
1
+ u
2
1r
_
u
2r
z
2
k
1
|u
1r
|
_
z
1
+ z
3
_
z
4
+ (1 + z
2
3
)
tan(
0r

1r
)
d
___
. (23)
Since z
1
, z
3
, and z
4
tend to zero, we can conclude
lim
t
_
d
dt
_
u
2
1r
z
1
_
u
2
1r
u
2r
z
2
_
= 0. (24)
Since u
2
1r
z
1
is uniformly continuous and converges to zero,
therefore, using Barbalats Lemma
d
dt
_
u
2
1r
z
1
_
also converges
to zero. Thus, using (24) and the assumption that u
1r
and u
2r
do not tend to zero from property 1, one obtains that z
2
tends
to zero. Consequently, the convergence of z
i
(i : 1 4) to zero
implies the convergence of V
1
to zero and asymptotic stability
of the closed-loop system.
Substituting of (18) into (16) yields kinematic control inputs
u
1
and u
2
. These kinematic control inputs will be assumed as
command signals for dynamic controller in next section, so we
represent them with u
c
= (u
1c
u
2c
).
IV. DYNAMIC CONTROLLER DESIGN
A. Dynamics Model
Mobile robot dynamics equations can be obtained using the
Lagrange method in matrix form as
M(q) q +C(q, q) = B(q) +A
T
(q) (25)
where M(q)
44
is the system inertia matrix, C(q, q)

41
is the centripetal and coriolis forces vector, B(q)
42
is the input transformation matrix, and A(q)
24
is the
system constraint matrix and is the vector of Lagrange multi-
pliers, [38]. The termA
T
(q) is the vector of constraint forces
exerted from the ground to the wheels, [5]. The system matrices
in (25) are dened as
M(q) =
_

_
M 0
0 M
Asin(
1
) Acos(
1
)
a
0
m
0
sin(
0
) a
0
m
0
cos(
0
)
A sin(
1
) a
0
m
0
sin(
0
)
Acos(
1
) a
0
m
0
cos(
0
)
I

1
a
0
d m
0
cos(
0

1
)
a
0
d m
0
cos(
0

1
) I

0
_

_
B(q) =
1
r
_
cos (
0
) sin (
0
) d sin (
0

1
) b
cos (
0
) sin (
0
) d sin (
0

1
) b
_
T
C(q, q) = [C
ij
]
41
= [
r

l
]
T
(26)
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
KEYMASI KHALAJI AND MOOSAVIAN: ROBUST ADAPTIVE CONTROLLER FOR A TRACTORTRAILER MOBILE ROBOT 5
where
C
11
= a
0
m
0
cos
0

2
0
Acos
1

2
1
C
21
= a
0
m
0
sin
0

2
0
Asin
1

2
1
C
31
= a
0
d m
0

2
0
sin(
0

1
)
C
41
= a
0
d m
0

2
1
sin(
0

1
)
M = m
0
+ m
1
; A = (a
1
m
1
+ dm
0
)
I

1
= m
1
a
2
1
+ m
0
d
2
+ I
1
; I

0
= m
0
a
2
0
+ I
0
.
Here, m
0
and m
1
denote the mass of the tractor and trailer
bodies respectively, I
0
and I
1
are their mass moments of inertia
about the vertical axis, a
0
and a
1
denote the distance between
center of masses and midpoint of wheels for tractor and trailer
bodies, respectively, and r is the radius of tractor wheels.
Substituting (2) into (25), and using the natural orthogonal
complement property given in (5), the array of Lagrange mul-
tipliers is removed and the system dynamic equations take the
following form:
M
1
(q) u +M
2
(q) u +

C(q, q) =

B(q) (27)
where
M
1
(q) =S
T
(q) M(q) S(q) , M
2
(q)=S
T
(q)M(q)

S(q)

C(q, q) =S
T
(q) C(q, q) ,

B(q) = S
T
(q) B(q) . (28)
Property 2: M
1
(q) is a symmetric and positive denite func-
tion.
Rearranging (27) we can write
=

B
1
(q)
_
M
1
(q) u +M
2
(q) u +

C(q, q)
_
. (29)
Equation (29) can be written as
= M

1
(q) u +M

2
(q) u +C

(q, q) (30)
where
M

1
(q) =

B
1
(q) M
1
(q) , M

2
(q) =

B
1
(q) M
2
(q)
C

(q, q) =

B
1
(q)

C(q, q) . (31)
B. Dynamics Model in the Presence of Uncertainties
A system dynamic model is an approximation of the real
system due to the presence of uncertainties and external distur-
bances. Therefore, the elements of the system dynamics model
in a real system can be dened in the following forms:
M

1
(q) =

M

1
(q) + M

1
(q)
M

2
(q) =

M

2
(q) + M

2
(q)
C

(q, q) =

C

(q, q) + C

(q, q) (32)
where

M

1
(q),

2
(q), and

C

(q, q) are the approximately

known parts, and M

1
(q), M

2
(q), and C

(q, q) are un-

known parts of the parameters M

1
(q), M

2
(q) and C

(q, q),
respectively.
Substituting these parameters into (30) yields
=

M

1
(q) u +

2
(q) u +

C

(q, q) + M

1
(q) u
+ M

2
(q) u + C

(q, q) +D (33)
where D
21
denotes the system disturbances vector.
Therefore, system uncertainties can be lumped as

1
= M

1
(q) u + M

2
(q) u + C

(q, q) +D. (34)

Substituting (34) into (33) one obtains
=

M

1
(q) u +

2
(q) u +

C

(q, q) +
1
. (35)
Hence, u can be calculated as
u =

M

1
1
(q) {

2
(q) u

C

(q, q)}
2
(36)
where

2
=

M

1
1
(q)
1
. (37)
Substituting (36) and (37) into (34), we obtain

1
= M

1
(q)

1
1
(q) {

2
(q) u

C

(q, q)
1
}
+M

2
(q) u + C

(q, q) +D. (38)

Simplications yield

1
= (I + M

1
(q)

1
1
(q))
1
M

1
(q)

1
1
(q)
{

2
(q) u

C

(q, q)}
+ (I + M

1
(q)

1
1
(q))
1
{M

2
(q) u + C

(q, q) +D}. (39)

Therefore, uncertainty vector
1
is upper-bounded as

1
(I + M

1
(q)

1
1
(q))
1
M

1
(q)

1
1
(q)
{ +

2
(q)u +

(q, q)}
+(I + M

1
(q)

1
1
(q))
1

{M

2
(q)u +C

(q, q) +D} (40)

where the Euclidean norm and spectral norm have been used for
vectors and matrices, respectively.
According to the nature of mechanical systems, the following
assumptions are considered.
1) The norm of system mass matrix is upper-bounded as
M(q) <
0
. (41)
2) The norm of S(q) matrix is upper-bounded as
S(q) <
1
. (42)
3) The norm of system unknown disturbances vector is
bounded as
D <
2
. (43)
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
6 IEEE/ASME TRANSACTIONS ON MECHATRONICS
4) The normof centripetal and coriolis forces vector is upper-
bounded as
C(q, q) <
3
+
4
q +
5
q
2
. (44)
Using (2) and (42), we can write (44) as
C(q, q) <
3
+
4
q +
5
u
2
(45)
where
5
=
5

2
1
.
5) The input transformation matrix is upper-bounded as
follows:
B(q) <
6
. (46)
6) In robotic systems, the following condition for the norm
of actuator inputs upper-bound is more realistic
<
7
+
8
q +
9
q
2
. (47)
Using (2) and (42), we can write (47) as
<
7
+
8
q +
9
u
2
(48)
where
9
=
9

2
1
.
According to the characteristics of robotic systems, these
assumptions are reasonable and mentioned bounded properties
have been used by many researchers [34], [39].
Considering the assumptions and using (31), following rela-
tions can be concluded
(I + M

1
(q)

1
1
(q))
1
<
10
M

1
(q)

1
1
(q) <
11
M

2
(q) <
12

S, M

1
(q) <
13
. (49)
Therefore, the norm of the uncertainty vectors from (40) and
(37) are upper-bounded as

1
<
0
+
1
q +
2
u
2
+
3

Su

2
<
0
+
1
q +
2
u
2
+
3

Su (50)
where
i
(i = 0 3) and
i
(i = 0 3) are upper-bounds pa-
rameters for system uncertainties.
V. FEEDBACK LINEARIZING DYNAMIC CONTROLLER
In this section, an FLDC is proposed for the control of the
WMR at the dynamic level. Dening the following tracking
error signal:
E = u
c
u (51)
where u
c
is the command velocity vector that was found in the
kinematic controller design. The aim of this section is to design
the actuator torques for the mobile robot to stabilize the tracking
errors around the origin.
Now, consider the following control law:
= M

2
(q) u +M

1
(q) u
c
+C

(q, q) +M

1
(q) KE
(52)
where K= diag (k
1
, k
2
) is a diagonal matrix with positive di-
agonal elements.
Proposition 2: The control law (52) for the dynamic model
(30) makes the origin of the tracking error vector (51) uniformly
asymptotically stable.
Proof: Consider the following positive denite function as a
Lyapunov candidate function
V
2
=
1
2
E
T
E. (53)
The time derivative of this function can be calculated as

V
2
= E
T

E = E
T
( u
c
u) . (54)
Using (30), we obtain

V
2
= E
T
( u
c
M

1
1
(q){ M

2
(q)u C

(q, q)}). (55)

Substituting (52) into (55) yields

V
2
= E
T
KE. (56)
The time derivative of the chosen Lyapunov candidate func-
tion is negative denite, hence with the designed control law
the tracking errors will converge to the origin and consequently,
WMR will asymptotically follow command signals.
VI. ROBUST ADAPTIVE FEEDBACK LINEARIZING
DYNAMIC CONTROLLER
In previous section, the feedback linearization controller was
designed as an ideal algorithm when perfect modeling is as-
sumed. In real applications, the system characteristics will devi-
ate from its ideal model due to uncertainties and external distur-
bances. In order to apply robust controllers for robotic systems
prior knowledge of upper-bounds corresponding to uncertain-
ties and external disturbances is essential. However, determi-
nation of these bounds is not easy due to the complexity and
unpredictability of the structure of uncertainties in dynamics of
robotics systems. The robust controllers that use prior knowl-
edge of uncertainties are designed for the worst case problem
and usually are conservative. This is due to the fact that such
controllers will not be fast enough and the entire capacity of
actuators will not be used in most of the time. To overcome
this problem, an adaptive control law is developed to estimate
upper-bounds of uncertainties of TTWR, and so an RAFLDC is
developed for the TTWR in this section.
The following RAFLDC law:
= M

2
(q) u +M

1
(q) u
c
+C

(q, q) +M

1
(q) KE
+M

1
(q)sgn(E)(
0
+
1
q +
2
u
2
+
3

Su)
(57)
with the following adaptive rules:

0
=
0
E ,


1
=
1
q E ,


2
=
2
u
2
E

3
=
3

S u E (58)
are developed for the dynamic system (35), where sgn (.) repre-
sents the signum function and
i
(i = 0 3) denote estimates
of modeling uncertainty upper-bound parameters. The estima-
tion error of these parameters can be dened as follows:
= . (59)
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
KEYMASI KHALAJI AND MOOSAVIAN: ROBUST ADAPTIVE CONTROLLER FOR A TRACTORTRAILER MOBILE ROBOT 7
Fig. 3. Proposed control law for the TTWR.
Proposition 3: The RAFLDC law (57) with the adaptation
rules (58) for the dynamic system (35) ensures the asymptotic
convergence of tracking errors to the origin in presence of un-
certainties and external disturbances.
Proof: Assume the following Lyapunov function candidate:
V
3
(E,

Q) =
1
2
(E
T
E +

Q
T

Q) (60)
where

Q=[

0

1

2

3 ]
T
, =diag
_

1
0
,
1
1
,
1
2
,
1
3
_
.
(61)
The time derivative of function V
3
_
E,

Q
_
can be calculated
as

V
3
(E,

Q) = E
T

E +

Q
T

Q = E
T
( u
c
u) +

Q
T

Q. (62)
Using (36), we obtain

V
3
(E,

Q) =

Q
T

Q
+ E
T
( u
c

1
1
(q){

2
(q) u

C

(q, q)} +
2
).
(63)
Substitution from (57) and (61), we can conclude

V
3
(E,

Q) = E
T
KE +E
T

2
+

Q
T

Q
E (
0
+
1
q +
2
u
2
+
3

S u). (64)
Consequently,

V
3
(E,

Q) E
T
KE +E
2
+
3

i=0

i


i

i
E (
0
+
1
q +
2
u
2
+
3

S u). (65)
Using (50) and (59), we will have

V
3
(E,

Q) E
T
KE +
0
_


0

0
E
_
+
1
_


1

1
q E
_
+
2
_


2

2
u
2
E
_
+
3
_


3

3
E
_
_
_

S
_
_
_ u
_
. (66)
TABLE I
SYSTEM PARAMETERS
Applying adaptive rules (58), we obtain

V
3
_
E,

Q
_
E
T
KE. (67)
By choosing appropriate gain matrix,

V
3
(E,

Q) will be neg-
ative semidenite. Consequently, E(t) and

Q are globally
bounded and also V
3
(E,

Q) will be globally bounded. There-
fore, tracking errors using the Barbalats Lemma asymptotically
converges to the origin.
In order to remove the chattering phenomenon, we may use
a continuous nonlinear function such as saturation function in-
stead of a hard nonlinear signum function [25].
VII. OBTAINED RESULTS
In this section, the proposed control law is applied on the
TTWR, and obtained results will be discussed. The overall
closed-loop control diagram for TTWR is shown in Fig. 3.
A. Experimental Setup
The developed TTWR system includes a differentially driven
vehicle towing a trailer as shown in Fig. 1.The tractor is equipped
with two differential driving wheels while a spherical wheel is
used for stable motion, and the trailer has two passive wheels.
System mass and geometric parameters are given in Table I.
The actuated wheels are driven by dc servo-motors that have
1.62 N.m holding torque, and 12 V operating voltage. The servo
motors are controlled by the PWM signals. We employ an I/O
card to generate PWM signals, and to drive the dc servo mo-
tors. A real-time image processing module is used as the mea-
surement system that obtains the posture information of the
TTWR by installing a camera above the motion plane; see
Fig. 1. This module consists of a camera with resolution of
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
8 IEEE/ASME TRANSACTIONS ON MECHATRONICS
TABLE II
CONTROLLER PARAMETERS
640 480 pixels and frame rate of 30 frames/s, and with a USB
connection transmits the gathered data to the host computer. The
frame rate of the camera is an important factor in real-time im-
age processing. Posture information is obtained using colored
labels on the TTWR, assuming an inertial frame attached to the
center of the motion plane visible for the camera. In order to
estimate the position of a specic colored label relative to an
inertial frame, the camera calibration process is performed. Sub-
sequently, image restoration has been done in order to remove
colored areas larger or smaller than colored labels. In order to
have a simple and fast algorithm, the motion of colored points
in the image sequence is followed. Therefore, position of point
P
0
can be obtained as
x
P 0
=
x
w
+ x
b
2
, y
P 0
=
y
w
+ y
b
2
(68)
where the indices w and b are specied to the white and
blue colors, respectively, and x
w
denotes the x position of
the center of white label area, and so on. Also, position
of P
1
(the reference point on the trailer) is directly mea-
sured as the position of the center of red label area, (x
r
, y
r
).
Therefore, system conguration parameters are obtained as
x = x
r
, y = y
r
,
1
= Atan2 (y
P 0
y
r
, x
P 0
x
r
), and
0
=
Atan2 (x
w
x
b
, y
b
y
w
) , where the index r is specied to
the red color.
A Laptop (32 bit, Intel Core 2 CPU 2.00 GHz with 2.00 GB
RAM) with USB connections to the camera is used for image
processing, control, and data transmissions. The controller is
implemented using MATLAB/Simulink. In experimental sys-
tems, lower limit for sampling time is often restricted to the
characteristics of the experimental systems hardware (sensors,
processors, communication devices, actuators, and so on). The
upper limit for sampling time is usually restricted to perfor-
mance and stability of the closed-loop system. Based on these
limitations, sampling time for the system should be selected.
For our experimental system, the lower limit is restricted to the
33 ms for the vision measurement system, and the upper limit
is about 0.11 s that leads to the instability of the closed loop
system as empirically determined.
B. Case Studies
In this section, several case studies have been done in order to
evaluate the effectiveness of the proposed controller. Controller
parameters are given in Table II. The reference trajectories are
considered as follows:
x
r
(t) = 0.02
_
R + cos
_
36t
T
__
cos
_
6t
T
_
y
r
(t) = 0.02
_
R + cos
_
36t
T
__
sin
_
6t
T
_
.
Fig. 4. Desired path for the system reference point in planar motion.
The desired path for the system reference point (P
1
) in its
planar motion in the Cartesian space is shown in Fig. 4.
The initial conditions of the system variables are assumed as
x(0) = 0.35, y (0) = 0.58,
1
(0) = 0.4 and

0
(0) = 0.4. (69)
It should be noted that in real implementation it is needed
to measure system variables in every instant of time. These
variables contain systemconguration vector qin Cartesian task
space and systemvelocity vector u. Real-time image processing
is used to measure these variables.
1) Robustness to Disturbances and Uncertainties: In this
section, an FLDC and the proposed RAFLDC are implemented
on the developed TTWR under the effects of external distur-
bances, parameter variations, and saturated inputs. The system
dynamics parameters uncertainties such as variations of masses
and moments of inertia in loading and unloading the TTWR are
practically expected.
In this case, the mobile robot is disturbed by periodic external
disturbances exerted starting from 25th second as
D = 0.65
_
sin
_
t
5
_
sin
_
t
15
__
T
U(t 25)
where U(t) is the unit step function. It is also assumed that
saturation limit of the actuator torques is 1 N.m, and systemmass
parameters are considered to be deviated from their nominal
values in 40th second as g (1 + 0.5U(t 40))g, where g
represents system parameters as g {m
0
, m
1
}.
The systemupper-bound parameters are estimated online and
shown in Fig. 5. Tracking errors are measured using image pro-
cessing module are shown in Fig. 6. Also dynamic control inputs
are shown in Fig. 7, which are less demanding for the proposed
RAFLDC. This is a consequence of appropriate selection of
controller gains in order to avoid reaching to saturation limits
of the actuators.
As can be seen from (57) and (52), choosing very low val-
ues for adaptive gains converts the RAFLDC to FLDC and the
results presented for FLDC will be obtained. In other words,
the robustness of the controller will diminishes if the adaptive
gains set to very low values. On the other hand very large values
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
KEYMASI KHALAJI AND MOOSAVIAN: ROBUST ADAPTIVE CONTROLLER FOR A TRACTORTRAILER MOBILE ROBOT 9
Fig. 5. Estimated parameters under uncertainties and saturated inputs for the
RAFLDC algorithm.
Fig. 6. Tracking reference trajectories under uncertainties and saturated inputs
for RAFLDC and FLDC. (a) Tracking of x-coordinate. (b) Tracking of y-
coordinate. (c) Tracking of
1
-coordinate. (d) Tracking of
0
-coordinate.
for adaptive gains besides increasing the rate of convergence of
the parameters (adaptation speed) causes greater control inputs
that may violate the saturation limits of the actuators. There-
fore, in order to maintain the stability of the closed-loop system
and also the robustness of the controller, we need to compro-
mise these effects. Hence, appropriate values for adaptive gains
should be chosen based on the characteristics and limitations of
the experimental system.
Fig. 7. Dynamic control inputs under uncertainties and saturated inputs for
RAFLDC and FLDC. (a) Right wheel actuator torque. (b) Left wheel actuator
torque.
Obtained results show that the proposed RAFLDC acts better
than a perfect FLDC, dealing with uncertainties and external
disturbances. The proposed algorithm reveals the characteris-
tic of nite time convergence of upper-bound parameters and
tracking errors under the effect of large time-varying periodic
disturbances, uncertainties, and saturated inputs. It is seen that
the proposed RAFLDC successfully eliminates the effects of
uncertainties, and external disturbances in a real application.
Obtained results indicate that the proposed algorithm is remark-
ably effective, which may be employed for other nonholonomic
systems.
2) Large Initial Orientation Mismatches: In this section, the
FLDC and the proposed RAFLDC are implemented on the de-
veloped TTWR with large initial mismatches, while the effects
of aforementioned external disturbances, parameter variations,
and saturated inputs are also included.
The initial conditions of system variables are assumed as
x(0) = 0.35, y (0) = 0.58,
1
(0) = , and

0
(0) = 0.4 (70)
which compared to (69) corresponds to a large mismatch be-
tween the tractor and trailer orientations.
The desired path for the system reference point (P
1
) in its
planar motion in the Cartesian space is the same as before as
shown in Fig. 4, and tracking errors are also shown in Fig. 8. It
can be seen that the proposed RAFLDC is capable of compen-
sating uncertainties and attenuating external disturbances much
better than FLDC, in the presence of parameter deviations and
model uncertainties and also large initial mismatches.
3) Robustness to Unmodeled Dynamics Case: In this paper,
it is assumed that the system has parametric uncertainties and
external disturbances such as variations of masses and moments
of inertia in loading and unloading the TTWR. The proposed
controller is also able to slightly alleviate the effects of unmod-
eled dynamics such as kinematic perturbations due to off-axle
hitching, though it cannot completely reject these uncertainties.
Designing control algorithms that can compensate unmodeled
dynamics is also a challenging problem that needs more investi-
gations, and makes the trend of our future works. In order to test
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
10 IEEE/ASME TRANSACTIONS ON MECHATRONICS
Fig. 8. Tracking reference trajectories under uncertainties and saturated in-
puts and large initial mismatches for RAFLDC and FLDC. (a) Tracking of
x-coordinate. (b) Tracking of y-coordinate. (c) Tracking of
1
-coordinate.
(d) Tracking of
0
-coordinate.
the performance of the controller, simulations have been done
on a TTWR with off-axle hitching as shown in Fig. 2.
The kinematic model of the TTWR with off-axle hitching is
calculated as follows:
_

_
x = u
1
cos
1
y = u
1
sin
1

1
=
1
d
1
{u
1
tan(
0

1
) d
0
u
2
sec(
0

1
)}

0
= u
2.
(71)
The off-axle hitching is assumed as d
0
= 4 cm in simulation
studies as an unmodeled dynamics effect. Considering the sys-
tem geometric dimensions, for instance the radius of the wheels
are equal to 2.6 cm, this off-axle hitching is expected to in-
troduce a high effect. The desired path for the system reference
point (P
1
) in its planar motion in the Cartesian space is the same
as before, and tracking errors are also shown in Fig. 9. These
results show that the proposed controller slightly attenuates the
effects of unmodeled dynamics such as kinematic perturbations
due to off-axle hitching, though it cannot completely reject this
uncertainty. It should be noted that the off-axle hitching of d
0
=
4 cm is not introduced to the controller, while assumed to be ex-
istent in the real system. Obviously, such magnicent mismatch
Fig. 9. Tracking errors in presence of off-axle hitching for RAFLDC and
FLDC. (a) Tracking error of x-coordinate. (b) Tracking error of y-coordinate.
(c) Tracking error of
1
-coordinate. (d) Tracking error of
0
-coordinate.
between the dynamics models (used in the controller and the
system itself) is rarely expected in practice.
VIII. CONCLUSION
In this paper, a newRAFLDCwas developed for the trajectory
tracking problem of a TTWR that is a highly nonlinear, under-
actuated, nonholonomic, and modular robotic system. First, the
system dynamic model was derived. Next, feasible reference
trajectories were generated and a stable kinematic controller
based on the Lyapunov approach was presented for the trajectory
tracking problem of the TTWR. A structure for upper-bounds
of uncertainties for the TTWR was analytically obtained. A ro-
bust adaptive controller at the dynamic level was proposed in the
presence of uncertainties and external disturbances. Appropriate
adaptive rules were developed to estimate parameters of upper-
bounds of uncertainties. Therefore, prior knowledge of bounds
of uncertainties is not required. The stability of the control algo-
rithm was veried using the Lyapunov method. Therefore, the
convergence of upper-bounds parameters and tracking errors
in nite time is guaranteed. Both simulation and experimental
results reveal the merits of the proposed controller in terms of
its robustness to uncertainties and external disturbances, even in
the presence of large initial mismatch between the tractor and
trailer orientations, and unmodeled dynamics effects.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
KEYMASI KHALAJI AND MOOSAVIAN: ROBUST ADAPTIVE CONTROLLER FOR A TRACTORTRAILER MOBILE ROBOT 11
REFERENCES
[1] S. A. A. Moosavian, A. Kalantari, H. Semsarilar, E. Aboosaeedan, and
E. Mihankhah, ResQuake: A tele-operative rescue robot, J. Mech. De-
sign, vol. 131, 2009.
[2] M. Eslamy and S. A. A. Moosavian, Dynamics and cooperative object
manipulation control of suspended mobile manipulators, J. Intell. Robot.
Syst., vol. 60, pp. 181199, 2010.
[3] S. A. A. Moosavian and A. Pourreza, Heavy object manipulation by a
hybrid serialparallel mobile robot, Int. J. Robot. Autom., vol. 25, 2010.
[4] S. Russo, K. Harada, T. Ranzani, L. Manfredi, C. Stefanini, A. Menciassi,
and P. Dario, Design of a robotic module for autonomous exploration
and multimode locomotion, IEEE/ASME Trans. Mechatronics, to be
published.
[5] G. Campion, G. Bastin, and B. Dandrea-Novel, Structural properties
and classication of kinematic and dynamic models of wheeled mobile
robots, IEEE Trans. Robot. Autom., vol. 12, no. 1, pp. 4762, Feb. 1996.
[6] P. Morin and C. Samson, Trajectory tracking for nonholonomic vehi-
cles, in Robot Motion and Control. vol. 335. London, U.K.: Springer,
2006, pp. 323.
[7] W. Danwei and X. Guangyan, Full-state tracking and internal dynamics of
nonholonomic wheeled mobile robots, IEEE/ASMETrans. Mechatronics,
vol. 8, no. 2, pp. 203214, Jun. 2003.
[8] A. Luca, G. Oriolo, and C. Samson, Feedback control of a nonholo-
nomic car-like robot, in Robot Motion Planning and Control., vol. 229,
J. P. Laumond, Ed. Berlin, Germany: Springer, 1998, pp. 171253.
[9] C. Altani, Some properties of the general n-trailer, Int. J. Control,
vol. 74, pp. 409424, 2001.
[10] Y. Nakamura, H. Ezaki, T. Yuegang, and C. Woojin, Design of steering
mechanism and control of nonholonomic trailer systems, IEEE Trans.
Robot. Autom., vol. 17, no. 3, pp. 367374, Jun. 2001.
[11] J. L. Martinez, J. Morales, A. Mandow, and A. Garcia-Cerezo, Steering
limitations for a vehicle pulling passive trailers, IEEE Trans. Control
Syst. Technol., vol. 16, no. 4, pp. 809818, Jul. 2008.
[12] J. P. Laumond, Controllability of a multibody mobile robot, IEEETrans.
Robot. Autom., vol. 9, no. 4, pp. 755763, Dec. 1993.
[13] J. P. Laumond, S. Sekhavat, and F. Lamiraux, Guidelines in nonholo-
nomic motion planning for mobile robots, in Robot Motion Planning and
Control., vol. 229, J. P. Laumond, Ed. Berlin, Germany: Springer, 1998,
pp. 153.
[14] F. Lamiraux and J. P. Laumond, Flatness and small-time controllability of
multibody mobile robots: Application to motion planning, IEEE Trans.
Autom. Control, vol. 45, no. 10, pp. 18781881, Oct. 2000.
[15] P. Rouchon, M. Fliess, J. L

EVine, and P. Martin, Flatness and motion

planning: the car with n trailers, in Proc. 2nd Eur. Control Conf., Gronin-
gen, The Netherlands, 1993, pp. 15181522.
[16] K. Matsushita and T. Murakami, Nonholonomic equivalent disturbance
based backward motion control of tractor-trailer with virtual steering,
IEEE Trans. Ind. Electron., vol. 55, no. 1, pp. 280287, Jan. 2008.
[17] K. Tanaka, S. Hori, and H. O. Wang, Multiobjective control of a vehi-
cle with triple trailers, IEEE/ASME Trans. Mechatronics, vol. 7, no. 3,
pp. 357368, Sep. 2002.
[18] J. H. Lee, W. Chung, M. Kim, and J.-B. Song, A passive multiple trailer
system with off-axle hitching, Int. J. Control, Autom., Syst., vol. 2,
pp. 289297, 2004.
[19] P. Bolzern, R. M. DeSantis, A. Locatelli, and D. Masciocchi, Path-
tracking for articulated vehicles with off-axle hitching, IEEE Trans.
Control Syst. Technol., vol. 6, no. 4, pp. 515523, Jul. 1998.
[20] M. Sampei, T. Tamura, T. Kobayashi, and N. Shibui, Arbitrary path
tracking control of articulated vehicles using nonlinear control theory,
IEEE Trans. Control Syst. Technol., vol. 3, no. 1, pp. 125131, Mar. 1995.
[21] A. Astol, P. Bolzern, and A. Locatelli, Path-tracking of a tractor-trailer
vehicle along rectilinear and circular paths: ALyapunov-based approach,
IEEE Trans. Robot. Autom., vol. 20, no. 1, pp. 154160, Feb. 2004.
[22] A. W. Divelbiss and J. T. Wen, Trajectory tracking control of a car-trailer
system, IEEE Trans. Control Syst. Technol., vol. 5, no. 3, pp. 269278,
May 1997.
[23] J. Roh and C. Woojin, Reversing control of a car with a trailer using a
driver assistance system, in Proc. IEEE Workshop Advanced Robot. Soc.
Impacts, 2010, pp. 99104.
[24] J.-C. Ryu, S. K. Agrawal, and J. Franch, Motion planning and control
of a tractor with a steerable trailer using differential atness, J. Comput.
Nonlinear Dyn., vol. 3, p. 031003, 2008.
[25] J. J. Slotine and W. Li, Applied Nonlinear Control. Englewood Cliffs,
NJ, USA: Prentice-Hall, 1991.
[26] E. D. Engeberg and S. G. Meek, Adaptive sliding mode control for pros-
thetic hands to simultaneously prevent slip and minimize deformation
of grasped objects, IEEE/ASME Trans. Mechatronics, vol. 18, no. 1,
pp. 376385, Feb. 2013.
[27] H. Chuxiong, Y. Bin, and W. Qingfeng, Performance-Oriented adap-
tive robust control of a class of nonlinear systems preceded by unknown
dead zone with comparative experimental results, IEEE/ASME Trans.
Mechatronics, vol. 18, no. 1, pp. 178189, Feb. 2013.
[28] Z. Chen, B. Yao, and Q. Wang, Accurate motion control of linear mo-
tors with adaptive robust compensation of nonlinear electromagnetic eld
effect, IEEE/ASME Trans. Mechatronics, vol. 18, no. 3, pp. 11221129,
Jun. 2013.
[29] R. Marino and P. Tomei, Nonlinear Control Design: Geometric, Adap-
tive, and Robust (Prentice Hall Information and System Sciences Series).
London, U.K.: Prentice-Hall, 1995.
[30] Z. Li, S. S. Ge, M. Adams, and W. S. Wijesoma, Robust adaptive control
of uncertain force/motion constrained nonholonomic mobile manipula-
tors, Automatica, vol. 44, pp. 776784, 2008.
[31] H. Zeng-Guang, Z. An-Min, C. Long, and T. Min, Adaptive control of
an electrically driven nonholonomic mobile robot via backstepping and
fuzzy approach, IEEE Trans. Control Syst. Technol., vol. 17, no. 4,,
pp. 803815, Jul. 2009.
[32] L. Cheng, Z.-G. Hou, and M. Tan, Adaptive neural network tracking
control for manipulators with uncertain kinematics, dynamics and actuator
model, Automatica, vol. 45, pp. 23122318, 2009.
[33] V. Nekoukar and A. Erfanian, Adaptive fuzzy terminal sliding mode
control for a class of MIMO uncertain nonlinear systems, Fuzzy Sets
Syst., vol. 179, pp. 3449, 2011.
[34] Z. Man and X. Yu, Adaptive terminal sliding mode tracking control for
rigid robotic manipulators with uncertain dynamics, JSMEInt. J., vol. 40,
pp. 493502, 1997.
[35] A. Isidori, Nonlinear Control Systems. New York, NY, USA: Springer,
1995.
[36] J. Levine, Analysis and Control of Nonlinear Systems: A Flatness-based
Approach. New York, NY, USA: Springer, 2010.
[37] H. K. Khalil, Nonlinear Systems. Englewood Cliffs, NJ, USA: Prentice-
Hall, 2002.
[38] J. H. Ginsberg, Engineering Dynamics. Cambridge, U.K.: Cambridge
Univ. Press, 2008.
[39] C. Abdallah, D. M. Dawson, P. Dorato, and M. Jamshidi, Survey of robust
control for rigid robots, IEEE Control Syst., vol. 11, no. 2, pp. 2430,
Feb. 1991.
Ali Keymasi Khalaji received the B.Sc. degree from
Iran University of Science and Technology, Tehran,
Iran, and the M.Sc. degree from K. N. Toosi Uni-
versity of Technology (KNTU), Tehran, in 2007 and
2009, respectively, where he is currently working to-
ward the Ph.D. degree in the Department of Mechan-
ical Engineering.
His research interests include modeling and con-
trol of mechanical systems, nonlinear control, adap-
tive and robust control with applications to mobile
robotics systems, and mechatronics.
S. Ali A. Moosavian received the B.S. degree
from Sharif University of Technology, Tehran, Iran,
the M.S. degree from Tarbiat Modaress University,
Tehran, and the Ph.D. degree from McGill Univer-
sity, Montreal, QC, Canada, in 1986, 1990, and 1996,
respectively, all in mechanical engineering.
He is currently a Professor with the Department
of Mechanical Engineering, K. N. Toosi Univer-
sity of Technology (KNTU), Tehran, where he has
been since 1997. He teaches courses in the areas of
robotics, dynamics, automatic control, analysis, and
synthesis of mechanisms. His research interests include the areas of dynamics
modeling and motion/impedance control of terrestrial, legged, and space robotic
systems. He has published more than 200 articles in peer-reviewed journals and
conference proceedings. He is one of the founders of the ARAS Research Group,
and the Director of the Center of Excellence in Robotics and Control at KNTU.