A Comparison of Adaptive Trajectory Tracking

Controllers for Wheeled Mobile Robots

Ehab I. Al Khatib1, Wasim M. F. Al-Masri1, Shayok Mukhopadhyay2, Mohammad A. Jaradat3, 4, and Mamoun Abdel-Hafez3
1
Mechatronics Engineering Graduate Program, American University of Sharjah, Sharjah, UAE.
2
Department of Electrical Engineering, American University of Sharjah, Sharjah, UAE.
3
Department of Mechanical Engineering, American University of, Sharjah, Sharjah, UAE.
4
Department of Mechanical Engineering, Jordan University of Science and Technology, Irbid, Jordan.
[email protected], [email protected], [email protected], [email protected], [email protected]

Abstract — Two adaptive trajectory tracking controllers for the dynamic model, and a term is added to the update law to
wheeled mobile robots are tested in this work. Adaptively tuned avoid parameter drift. In [6], the authors propose a method
proportional control is one approach, where as the other for designing adaptive controllers which can be used with
controller uses a Universal Adaptive Stabilization (UAS) based mobile robots whose parameters are unknown. In [7], a
technique. Using simulations, the robustness of the above
controllers is quantified in the presence of measurement noise.
piecewise smooth controller is used for a mobile with two
The robustness is measured in terms of the Integral of absolute degrees of freedom, and the well-known problem of
magnitude of the error (IAE), the Integral of square of the error chattering is avoided. Beside the problem of control of
(ISE), and the Integral of time multiplied by the absolute value mobile wheeled robot, Many studies has been done on
of the error (ITAE) criteria. It is observed that the UAS based adaptive controllers based on Nussbaum gain method, in [8],
technique shows fast convergence in the absence of noise. To a new adaptive control scheme for nonlinear multivariable
combat the effect of noise, the authors reset the adaptation gains MRAC systems has proposed, based on the Nussbaum gain
after the adaptation gains reach a preset bound. With this method and the backstepping approach. A systematic
technique it is found that the UAS based technique converges to procedure is developed for designing global adaptive control
the trajectory being tracked faster than the adaptively tuned
proportional controller, and also faster than a traditional input-
of a class of nonlinear systems in [9], the authors study a class
output state feedback linearization based controller. of systems possesses a triangular structure and can be of
arbitrary dynamic order without a priori knowledge of the
I. INTRODUCTION signs of certain parameters and without growth restriction.
Wheeled mobile robots (WMRs) are common in service
In this paper, an adaptive trajectory-tracking controller is
robotics. Several configurations of WMRs exist. Some of the
proposed based on robot kinematics, and its stability is
common configurations are differential drive robots,
proved. Two adaptive controllers are presented. The first one
omnidirectional wheeled robots. A detailed study of the
is an adaptive proportional controller (APC), which
kinematics of WMRs is found in [1]. Perfect rolling
adaptively tunes a proportional controller's gains. The second
constraints for a robot's wheels (i.e. no lateral slippage) [1,
one is a Nussbaum based controller (NC), which is an
2], and nonholonomic motion constraints preventing
adaptive controller which uses a Nussbaum function based
sideways robot motion add to the challenges of motion
strategy for adaptive stabilization [10].
planning with WMRs.
This paper is organized as follows. In Section II, a dynamic
Approaches toward trajectory tracking using WMRs may be
model of WMR is derived. Section III presents the control
classified in two main types [3]-[6]. The first type focuses
design. Section IV presents the simulation results and
only on robot kinematics. The second type uses controllers
conclusions are presented in Section V.
which consider both the robot kinematics and robot
dynamics. In [3], the authors study the motion control
problem of wheeled mobile robots (WMRs) in environments II. KINEMATIC MODEL FOR WHEELED MOBILE ROBOT

without obstacles. The authors consider the popular unicycle The derivations of the kinematics of the mobile robot used in
kinematics, it is shown that dynamic feedback linearization this paper are based on the assumption that the robot is made
is an efficient design tool leading to a solution up of a rigid frame equipped with nondeformable wheels and
simultaneously valid for both trajectory tracking and setpoint that it is moving on a horizontal plane.
regulation problems. Control strategies for wheeled mobile
robots subjected to nonholonomic constraints is studied in [4] We provide some background information from [2]. In order
based on the kinematic and dynamic equations of motion and to specify the position of the robot on the plane, we establish
the actuator dynamics. a relationship between the global reference frame on the
plane and the local reference frame on the robot. These
Adaptive control is attractive for situations where robustness frames are shown in figure 1. The axes ܺூ and ܻூ define an
to variation in system parameters is required. An adaptive arbitrary inertial basis on the plane as a global reference
trajectory-tracking controller for a unicycle-like mobile robot frame from an originܱ. To specify the position of the robot,
was designed and fully tested in [5]. This controller is divided choose a point ܲ on the robot chassis as its position reference
in two parts, which are based on the kinematic and dynamic point. The basis ሼܺோ ǡ ܻோ ሽdefines two axes passing through ܲ
models of the robot. The model considered in [5] takes the on the robot chassis and is the robot’s local reference frame.
linear and angular velocities as input reference signals. The position ofܲ in the global reference frame is specified
Further, an adaptive parameter update law is used in [5] with
978-1-4673-7797-3/15/$31.00 ©2015 IEEE
by coordinates ‫ ݔ‬and‫ݕ‬, and the angular difference between The combination of wheel rolling and sliding constraints for
the global and local reference frames (i.e. the angle between all wheels of the robot describes the kinematic behavior and
ܺோ and ܺூ ) is given byߠ. The position of the robot is then the resulting equation is [2]:
completely specified by the three variables‫ݔ‬ǡ ‫ݕ‬ǡ ߠ. We define
௥ఝሶభ ௥ఝሶ
a 3 x 1-vector ߦdescribing the robot posture where ߦூ with ൅ మ
ଶ ଶ
respect to global frame and ߦோ with respect to local frame, ߦூሶ ൌ ܴሺߠሻ ൦
ିଵ
൪ (7)
Ͳ
and also define the orthogonal rotation matrix ܴሺߠሻ as ௥ఝሶభ ି௥ఝሶమ
follows. ൅
ଶ௟ ଶ௟
ߦூ ൌ ሾ‫ߠݕݔ‬ሿ் (1) The linear velocity (‫ݒ‬ሻ of the robot is always heading in the
ܿ‫Ͳ ߠ݊݅ݏ ߠݏ݋‬ ܺோ direction of the robot’s reference frame due to the
ܴሺߠሻ ൌ ൥െ‫Ͳ ߠݏ݋ܿ ߠ݊݅ݏ‬൩ (2) nonholonomic constraint and the angular velocity is (‫ )ݓ‬can
Ͳ Ͳ ͳ be considered as the rotation speed of the local reference
So ܴሺߠሻ defines the relation between ߦூሶ , and ߦோሶ as follows. frame with respect to the global reference frame. By writing
ߦோሶ ൌ ܴሺߠሻߦூሶ (3). (7) in terms of ‫ ݒ‬and ‫ݓ‬ǣ
‫ݔ‬ሶ
ߦூሶ ൌ ൥‫ݕ‬ሶ ൩
ߠሶ
௥ఝሶభ ௥ఝሶ
ܿ‫Ͳ ߠ݊݅ݏ ߠݏ݋‬ ିଵ ൅ మ ‫ߠݏ݋ܿݒ‬
ଶ ଶ
ൌ ൥െ‫Ͳ ߠݏ݋ܿ ߠ݊݅ݏ‬൩ ൦ Ͳ ൪ ൌ ൥ ‫ ߠ݊݅ݏݒ‬൩ (8)
௥ఝሶభ ି௥ఝሶమ
Ͳ Ͳ ͳ ൅ ‫ݓ‬
ଶ௟ ଶ௟
Where
௥ఝሶ ௥ఝሶ
‫ݒ‬ൌ భ൅ మ (9)
ଶ ଶ
௥ఝሶభ ௥ఝሶమ
Figure 1: The global reference frame and the robot local ‫ݓ‬ൌ െ  (10)
ଶ௟ ଶ௟
reference frame.
III. CONTROL DESIGN
This differential drive robot has two wheels, each with The control system for trajectory tracking using a WMR is
diameter ‫ ݎ‬. Given a point ܲ centered between the two drive made of a high-level controller which computes the linear
wheels, each wheel is at a distance ݈ fromܲ. Given ‫ݎ‬ǡ ݈ǡ and angular velocity then sends them to a low-level controller
ߠǡand the spinning speed of each wheel߮ଵሶ and߮ሶ ଶ , a as an input. The low-level controller, then controls the motors
forward kinematic model for the robot’s overall speed in the in a way that the desired linear and angular velocity is
global reference frame is: achieved. A step towards solving the trajectory tracking
ߦூሶ ൌ ሾ‫ݔ‬ሶ ‫ݕ‬ሶ ߠሶ ሿ் ൌ ݂ሺ݈ǡ ‫ݎ‬ǡ ߠǡ ߮ଵሶ ǡ ߮ଶሶ ሻǤ (4) problem involves Input-Output State Feedback Linearization
(IO-SFL). Two adaptive controllers are proposed further, the
first one is an adaptive proportional controller (APC), and the
second one is Nussbaum function based adaptive controller
(NC). The details of I-O SFL, APC, and the NC are provided
in the following subsections.
A. Input-Output State Feedback linearization (I-O SFL).
In this section input output state feedback linearization is
presented. By considering a point B (‫ݔ‬௕ ǡ ‫ݕ‬௕ ) outside the
wheel axle of the unicycle model with distance b, as a point
of reference for the WMR, this is shown in figure 3. It is
possible to control the robot motion with a constant linear
velocity regardless of the path curvature [11].

Figure 2: A fixed standard wheel and its parameters [2].

Before we present the derivation of the kinematic model of

the WMR, two constraints will be presented for every fixed
wheel. Figure 2 shows a wheel ‫ ܣ‬and describes its pose with
respect to local reference frame. Where ߙƒ†ߚ deterime the
orientation of the wheel with respect to the local frame. The
first constraint enforces the concept of rolling contact – i.e.
that the wheel must roll when motion takes place in the
appropriate direction, represented by (5). The second
constraint enforces the concept of no lateral slippage – i.e.
that the wheel must not slide orthogonal to the wheel plane,
represented by (6).
ሾ•‹ሺߙ ൅ ߚሻ  െ ‘•ሺߙ ൅ ߚሻ ሺെ݈ሻܿ‫ߚݏ݋‬ሿܴሺߠሻߦூሶ െ ‫߮ݎ‬ሶ = 0 (5)
ሾ ‘•ሺߙ ൅ ߚሻ•‹ሺߙ ൅ ߚሻ ݈‫ߚ݊݅ݏ‬ሿܴሺߠሻߦூሶ ൌ Ͳ (6)
Figure 3: Control problem description.
From Fig. 3 we have, ൌ ݁௫ ሺ‫ݔ‬ሶ ௗ௘௦ െ ‫ݔ‬ሶ ௕ ሻ ൅ ݇௫ ሺߛ௫ ݁௫ ଶ ሻ
‫ ܾݔ‬ൌ ‫ ݔ‬൅ ܾܿ‫ߠݏ݋‬ (11) ൌ ݁௫ ൫‫ݔ‬ሶ ௗ௘௦ െ ‫ݔ‬ሶ ௗ௘௦ െ ݇ଵ ଶ ݁௫ ൯ ൅ ݇ଵ ߛଵ ݁௫ ଶ
‫ ܾݕ‬ൌ ‫ ݕ‬൅ ܾ‫ߠ݊݅ݏ‬ (12) ଶ
ൌ െ݇ଵ ଶ ݁௫ ൅ ݇ଵ ߛଵ ݁௫ ଶ
The derivative of eq. (11), and eq. (12) are shown below. ଶ
ൌ െሺ݇ଵ ଶ െ ݇ଵ ߛଵ ሻ݁௫ < 0 (36)
‫ݔ‬ሶ ܾ ൌ ‫ ߠݏ݋ܿݒ‬െ ‫ߠ݊݅ݏܾݓ‬ (13)
Such that ݇௫ ሺͲሻ ൐ Ͳ , then
‫ݕ‬ሶ ܾ ൌ ‫ ߠ݊݅ݏݒ‬െ ‫ߠݏ݋ܾܿݓ‬ (14) ܸ௫ ሺͲǡͲሻ ൌ Ͳ
ߠሶ ൌ ‫ݓ‬ (15) ܸ௫ ൐ Ͳǡ ‫݁׊‬௫ ǡ ݇ଵ ് Ͳ
‫ݔ‬ሶ ܾ ܿ‫ ߠݏ݋‬െܾ‫ݒ ߠ݊݅ݏ‬ ܸ௫ሶ ൏ Ͳǡ ‫݁׊‬௫ ǡ ݇ଵ ് Ͳ if ݇ଵ ൐ ߛଵ
൤‫ݕ‬ሶ ൨ ൌ ቂ ቃቂ ቃ (16)
ܾ ‫ݓ ߠݏ݋ܾܿ ߠ݊݅ݏ‬ By following the same procedure for y position, the same
ܿ‫ ߠݏ݋‬െܾ‫ߠ݊݅ݏ‬ result can be obtained. Then, the system is asymptotically
ቚ ቚ ൌ ܾ ് Ͳ(17) stable. Therefore, the error always converges to zero and
‫ߠݏ݋ܾܿ ߠ݊݅ݏ‬
For the matrix in eq. (16) to be invertible, b should not equal trajectory tracking goal is achieved.
to zero as shown in eq. (17). Then the linear and angular
velocity is as given in eq. (18) and eq. (19). C. Nussbaum function based adaptive controller (NC)
‫ݒ‬ ܿ‫ ߠݏ݋‬െܾ‫ ߠ݊݅ݏ‬െͳ ‫ݔ‬ሶ ܾ The NC is based on a standard methodology for adaptive
ቂ ቃൌቂ ቃ ൤‫ݕ‬ሶ ൨ (18) control of systems with unknown control directions. In this
‫ݓ‬ ‫ߠݏ݋ܾܿ ߠ݊݅ݏ‬ ܾ study, the adaptation is done as follows. First, ‫ݔ‬ሶ ௕ , and ‫ݕ‬ሶ ௕ are
‫ݒ‬ ܿ‫ߠݏ݋‬ ‫ݔ ߠ݊݅ݏ‬ሶ ܾ chosen as shown in eq. (37) and eq. (38).
ቂ ቃ ൌ ቈെ ͳ ‫ߠݏ݋ܿ ͳ ߠ݊݅ݏ‬቉ ൤‫ݕ‬ሶ ൨ (19)
‫ݓ‬ ܾ
ܾ ܾ ‫ݔ‬ሶ ܾ ൌ ‫ݔ‬ሶ ݀݁‫ ݏ‬൅ ܰͳ ሺ݇ͳ ሻ݁‫ݔ‬ (37)
Given a trajectory (‫ݔ‬ௗ௘௦ ǡ ‫ݕ‬ௗ௘௦ ) as shown in Fig. (3), it is
‫ݕ‬ሶ ܾ ൌ ‫ݕ‬ሶ ݀݁‫ ݏ‬൅ ܰʹ ሺ݇ʹ ሻ݁‫ݕ‬ (38)
possible to find input ‫ݔ‬ሶ ܾ and ‫ݕ‬ሶ ܾ that guarantee asymptotic
ܰଵ , and ܰଶ are defined as follows.
tracking, this can be seen as follows.
ܰଵ ሺ݇ଵ ሻ ൌ ݁ ௞భ •‹ ݇ଵ (39)
‫ݔ‬ሶ ܾ ൌ ‫ݔ‬ሶ ݀݁‫ ݏ‬൅ ݇ଵ ݁‫ݔ‬ (20)
ܰଶ ሺ݇ଶ ሻ ൌ ݁ ௞మ •‹ ݇ଶ (40)
‫ݕ‬ሶ ܾ ൌ ‫ݕ‬ሶ ݀݁‫ ݏ‬൅ ݇ଶ ݁‫ݕ‬ (21) Where
Where ݇ሶଵ ൌ ߛଵ ݁௫ ଶ (41)
݁௫ ൌ ሺ‫ݔ‬ௗ௘௦ െ ‫ݔ‬஻ ሻ (22) ݇ሶଵ ൌ ߛଶ ݁௬ ଶ (42)
݁௬ ൌ ሺ‫ݕ‬ௗ௘௦ െ ‫ݕ‬஻ ሻ (23) The proof of stability is left for future work, an approach to
the proof can be found along the lines of theorem 4.2.1 in
By rearranging eq. (20) and eq. (21), the following equations [10].
are obtained.
݁ሶ ‫ ݔ‬൅ ݇ଵ ݁‫ ݔ‬ൌ Ͳ (24) IV. SIMULATION RESULTS
݁ሶ ‫ ݕ‬൅ ݇ଶ ݁‫ ݕ‬ൌ Ͳ (25) To test the performance of the presented controllers, some
By solving (24) and (25), we get. simulation tests in different scenarios were done using
݁௫ ൌ ܿଵ ݁ ି݇ͳ ௧ (26) MATLAB to track a square shaped trajectory, which is
݁௬ ൌ ܿଶ ݁ ି݇ʹ ௧ (27) generated using parametric equations by defining four points
For positive ݇ଵ and ݇ଶ at the corners {(0,0),(0,20),(20,20,),(20,0)}. The initial pose
ߦூ of the robot in the simulation is [0.3 0.3 0.0]T. As shown
Ž‹ ݁௫ ሺ‫ݐ‬ሻ ൌ Ͳ (28)
௧՜ஶ in Fig. 4, based on the error between the reference path and
Ž‹ ݁௬ ሺ‫ݐ‬ሻ ൌ Ͳ (29) the B point (from Fig. 3), by using eq. (19), the controller
௧՜ஶ
Then the system is asymptotically stable and the error always calculates‫ݒ‬, and ‫ ݓ‬which will be used in the robot model to
converges to zero. calculate the pose of robot ߦூ by using eq. (8). In test 1, the
simulated sensors measure exactly the ߦூ which is given by
B. Adaptive Proportional Controller (APC) eq. (8), on the other hand, the simulated sensors in test 2
The APC is based on the tuning controller’s gains (݇ଵ measure ߦூ with added noise as explained in later sections.
and ݇ଶ ) which depends on the error. And it’s derived as
follows. e V, w Mobile Robot x, y, θ
Ref.
EKF Path + Controller
Model
Adaptation law: -
݇ሶଵ ൌ ߛଵ ݁௫ ଶ (30)
݇ሶଶ ൌ ߛଶ ݁௬ ଶ (31) Simulated
Such that ߛଵ ,ߛଶ are positive constants. Then, the adaptation Sensors

laws are substituted to what has been picked for ‫ݔ‬ሶ ܾ , and ‫ݕ‬ሶ ܾ , Figure 4: Control sytem diagram.
i.e. in eq. 20 and eq. 21.
A. Test 1: Controllers’ Performance Without Noise
‫ݔ‬ሶ ܾ ൌ ‫ݔ‬ሶ ݀݁‫ ݏ‬൅ ݇ͳ ʹ ݁‫ݔ‬ (32)
ʹ As shown in Fig. 5, thick red line represent the reference
‫ݕ‬ሶ ܾ ൌ ‫ݕ‬ሶ ݀݁‫ ݏ‬൅ ݇ʹ ݁‫ݕ‬ (33)
trajectory which starts from (0, 0). In the same figure, the thin
To prove the stability in x position, candidate Lyapunov blue line, the dashed magenta line, and the green star line
function is defined as follows. represent the robot positions achieve using I-O SFL, APC,
ଵ
ܸ௫ ሺ݁௫ ǡ ݇ଵ ሻ ൌ ሺ݁௫ ଶ ൅ ݇ଵ ଶ ሻ (34) and NC controllers respectively. For I-O SFL controller, the
ଶ
The time derivative of eq. (34) and simplification steps are robot converges to the trajectory within 1.5 second, for APC
shown in the following equations. convergence is seen within 1.2 second, and for NC
ܸ௫ሶ ሺ݁௫ ǡ ݇ଵ ሻ ൌ ݁௫ ݁௫ሶ ൅ ݇ଵ ݇ሶଵ (35) convergence is obtained within 0.7 second. For all controllers
in this test, the steady state error is within ‫Ͳט‬ǤͲͳmeter. B. Test 2: Controllers’ Performance With Noise
Therefore, the performance of the three controllers are good, A zero mean white noise is introduced in this test. Where the
but NC shows the best performance followed by APC then I- noise is added to true states as follows.
O SFL, and these results are verified in Table 1 – Table 3. ߦ௠௘௔௦௨௘௥௘ௗ =ሾ‫ߠݕݔ‬ሿ் +ሾ‫ݒݓݓ‬ሿ் (44)
Fig. 6 shows that the gains ݇ଵ ǡ ƒ†݇ଶ for APC settle to a Where
constant values, 1.092 for ݇ଵ ǡand 1.072 for ݇ଶ Ǥ Fig.7 shows ™̱ܰሺͲǡ “ଶ ሻ (45)
the same for ݇ଵ ǡ ݇ଶ ǡ ܰଵ ǡ ƒ†ܰଶ for NC which settle to 1.071 ˜̱ܰሺͲǡ ” ଶ ሻ (46)
for ݇ଵ , and 1.061 for ݇ଶ , 2.56 for ܰଵ ǡ and 2.52 for ܰଶ Ǥ The authors have used the values 0.1 meter for q, and 0.0001
rad for r. The performance of the I-O SFL, APC and NC is
shown in Fig. 8 – Fig. 10. It can be seen from Fig. 8 the
performance of the three controllers are good at the
beginning, but with time the gains ݇ଵ ǡ ƒ†݇ଶ for APC and
݇ଵ ǡ ݇ଶ ǡ ܰଵ ǡ ܽ݊݀ܰଶ for NC keep increasing as shown in Fig. 9
and Fig. 10, in NC case the gains ݇ଵ ƒ†݇ଶ increase rapidly
to high value, therefore, the gains ܰଵ ǡ ƒ†ܰଶ have diverged.
Because of this the robot cannot keep tracking the trajectory
as shown in Fig. 8.
To solve this problem, the following solution is proposed, by
making a restriction on the growth of NC’s gains as
explained in the following:
ߛ ݁ ଶ ǡ݇ଵ ൏ ʹ
݇ሶଵ ൌ ൜ ଵ ௫ (47)
Ͳǡ݇ଵ ൒ ʹ
ଶ
ߛ ݁ ǡ݇ଶ ൏ ʹ
݇ሶଶ ൌ ൜ ଶ ௬ (48)
Ͳǡ݇ଶ ൒ ʹ
Figure 5: WMR in tracking, a comparison between the controllers.
In addition to (47), (48) the value݇ଵ , ݇ଶ are reset to a
constant i.e. 1 when the upper bound of 2 is reached. The
performance of the enhanced NC is shown in Fig. 11 and in
this case the robot succeeds to track the trajectory with a good
performance even better than APC performance. It can be
seen from Fig. 12, the gains are reset to specific value (e.g.
1) each time they increase beyond specific value (e.g. 2). For
I-O SFL controller, the robot converges to the trajectory
within 1.7 second, for APC convergence is seen within 1.5
second, and for NC convergence is obtained within 0.9
second. For I-O SFL, and APC controllers in test 2, the
steady state error is within ‫Ͳט‬ǤͲͷmeter. On the other hand,
steady state error is within ‫Ͳט‬ǤͲͺmeter for NC.

Figure 6: APC gains vs. time.

Figure 8: WMR in tracking, a comparison between the controllers.

Figure 7: NC gains vs. time.
 Figure 9: APC gains vs. time.

Figure 12: NC gains vs. time.

For further evaluation, the performance criteria are obtained

and shown in Table 1 – Table 3. The first one is the Integral
of square of the error (ISE), which is given as follows.
ஶ
ܵଵ ൌ ‫׬‬଴ ݁ ଶ ሺ‫ݐ‬ሻ݀‫ݐ‬ (49)
The second one is the Integral of absolute magnitude of the
error (IAE), which is as follows.
ஶ
ܵଶ ൌ ‫׬‬଴ ȁ‡ሺ‫ݐ‬ሻȁ݀‫ݐ‬ (50)
The third one is the Integral of time multiplied by the
absolute value of the error (ITAE), which is given as follows.
ஶ
ܵଷ ൌ ‫׬‬଴ ‫ݐ‬ȁ‡ሺ‫ݐ‬ሻȁ݀‫ݐ‬ (51)
These performance indices are consistent with previous
discussed results. Where in Test 1, NC has the best
performance. In Test 2 where the white noise is added,
regular I-0 SFL shows the best performance followed by
enhanced NC.
Figure 10: NC gains vs. time. Table 1: Integral of square of the Controllers' error for each test.
Controller ISE
type Test 1 Test 2
IO-SFL 2.9033 4.0059
APC 2.8923 9.3695
NC 2.8367 8.1403

Table 2: Integral of absolute magnitude of the Controllers' error for each

test.
Controller IAE
type Test 1 Test 2
IO-SFL 27.4705 31.8875
APC 27.4465 47.1527
NC 27.3051 44.7608

Table 3: Integral of time multiplied by the absolute value of the

Controllers' error for each test.
Controller ITAE
type Test 1 Test 2
IO-SFL 3.6359e+04 4.1993e+4
APC 3.6375e+04 6.9169e+4
Figure 11: WMR in tracking, a comparison between the NC 3.6402e+04 6.0717e+4
controllers.
 V. Conclusion: validation," in Control Systems Technology, IEEE Transactions on,
vol.10, no.6, pp.835-852, Nov 2002.
The problem of trajectory tracking is addressed in this paper. [4] Ilan Zohar, Amit Ailon, Raul Rabinovici, Mobile robot characterized
This paper proposes the two adaptive controllers to guide a by dynamic and kinematic equations and actuator dynamics:
WMR during trajectory tracking. The first approach is based Trajectory tracking and related application, Robotics and Autonomous
Systems, Volume 59, Issue 6, June 2011, Pages 343-353, ISSN 0921-
on adapting proportional gains. In this approach, the gains 8890.
are changed and adapted with the error to control the robot. [5] Felipe N. Martins, Wanderley C. Celeste, Ricardo Carelli, Mário
The second controller is based on a Nussbaum function based Sarcinelli-Filho, Teodiano F. Bastos-Filho, An adaptive dynamic
adaptation gains. In the first test, in the absence of noise, NC controller for autonomous mobile robot trajectory tracking, Control
has the best performance followed by APC, finally, I-O SFL. Engineering Practice, Volume 16, Issue 11, November 2008, Pages
1354-1363, ISSN 0967-0661,
On the other hand, when the performance of the controllers
[6] Fukao, T.; Nakagawa, H.; Adachi, N., "Adaptive tracking control of a
has been tested in the presence of measurement noise in test nonholonomic mobile robot," in Robotics and Automation, IEEE
2, in the first try of test 2, NC fails due to sensitivity to the Transactions on , vol.16, no.5, pp.609-615, Oct 2000
noise. But after applying the solution, NC shows immunity [7] De Wit, C.C.; Sordalen, O.J., "Exponential stabilization of mobile
to the noise. robots with nonholonomic constraints," in Automatic Control, IEEE
Transactions on , vol.37, no.11, pp.1791-1797, Nov 1992.
[8] Ying Zhou, Yuqiang Wu, Output feedback adaptive control of
multivariable nonlinear systems using Nussbaum gain method*,
References Journal of Systems Engineering and Electronics,Volume 17, Issue 4,
December 2006, Pages 829-835, ISSN 1004-4132.
[1] Campion, G.; Bastin, G.; D'Andrea-Novel, B., "Structural properties [9] Ye Xudong; Jiang Jingping, "Adaptive nonlinear design without a
and classification of kinematic and dynamic models of wheeled mobile priori knowledge of control directions," in Automatic Control, IEEE
robots," Robotics and Automation, 1993. Proceedings., 1993 IEEE Transactions on , vol.43, no.11, pp.1617-1621, Nov 1998.
International Conference on , vol., no., pp.462,469 vol.1, 2-6 May [10] A.Ilchmann, Non-identifier-based high-gain adaptive control. London:
1993. Springer-Verlag, 1993.
[2] Siegwart R, Nourbakhsh IR, Scaramuzza D. Introduction to [11] B. d’Andr´ ea Novel, G. Campion, and G. Bastin, “Control of
autonomous mobile robots. 2nd ed. Cambridge, Mass: MIT Press; nonholonomic wheeled mobile robots by state feedback linearization,”
2011. Int. J. Rob. Res., vol. 14, no. 6, pp. 543–559, Dec. 1995.
[3] Oriolo, G.; De Luca, A.; Vendittelli, M., "WMR control via dynamic
feedback linearization: design, implementation, and experimental