2014 Tenth International Conference on Intelligent Information Hiding and Multimedia Signal Processing

Adaptive Fuzzy CMAC Design for an Omni-Directional Mobile Robot

Ter-Feng Wu Hsu-Chih Huang

Department of Electrical Engineering Department of Electrical Engineering
National Ilan University National Ilan University
Ilan 26047, Taiwan Ilan 26047, Taiwan.
[email protected] [email protected]

Pu-Sheng Tsai Nien-Tsu Hu

Department of Electronic Engineering Integrated Logistical Support Center
China University of Science and Technology Chung-Shan Institute of Science and Technology
Taipei 11581, Taiwan Taoyuan 32599, Taiwan
[email protected] [email protected]

Zhi-Qiang Yang
Department of Electrical Engineering
National Ilan University
Ilan 26047, Taiwan.
[email protected]

Abstract—This work presents an adaptive fuzzy

cerebellar model articulation controller (AFCMAC) for I. INTRODUCTION
solving the tracking control problem for an omni- In recent years, much research has been performed into
directional mobile robot. First, fuzzy logic and CMAC control problems associated with nonlinear control systems
are combined, coupled with a triangular basis function [1]. Systems theory and the traditional feedback control
that is embedded in the hypercube receptive-field space theory have led to various control schemes that depend on
to yield non-constant differentiable basis functions, exact mathematical system dynamical models. However,
which simplify the complex structure and reduce the most of these approaches involve many uncertainties when
number of input space dimensions of CMAC. Both implemented in practice, which may be associated with
adaptive and control laws are developed to tune all of incompleteness of the model or several unresolvable
the control parameters online, accommodating disturbances. A well-developed control system must exhibit
high dynamic performance and robustness even when
uncertainty in tracking control for an omni-directional
implemented under complexity and uncertainty [2].
mobile robot. Hardware agencies and drive circuits are
Intelligent control theory is utilized in dynamic models
developed. This proposed AFCMAC is implemented in a
to eliminate uncertainties that are caused by disturbances.
high-performance field-programmable gate array
Systems that are based on the theory include neural
(FPGA) chip using a hardware/software co-design
networks (NNs)[3], fuzzy control of complex processes with
method and the Qsys design concept with a reusable
reference to human experience [4]. Drawing on Marr’s
user IP (Intellectual Property) core library. An omni-
model, Albus established a cerebellar model known as the
directional mobile robot is thus produced. Finally, the
“cerebella model articulation controller” (CMAC) [5]. A
trajectory tracking controllability of the robot along an
neural network that models the structure and function of the
elliptical path is experimented to evaluate the
human cerebellum, the CMAC exhibits motor behaviors
effectiveness of the proposed AFCMAC control scheme.
without using any complex algorithm; it is a table look-up
Keywords—cerebellar model articulation controller (CMAC); mechanism that comprises a series of function mapping.
fuzzy set; adaptive control; omni-directional mobile robot The cerebellar model has advantages over conventional
NNs; it is composed of simple theoretical structures, requires

978-1-4799-5390-5/14 $31.00 © 2014 IEEE 839

DOI 10.1109/IIH-MSP.2014.212
easy computation, exhibits fast learning and favorable local ym
generalization, approximates multiple non-linear systems,
and can be easily implemented in both hardware and T1
software. If the CMAC is merged into a fuzzy set, its T2

complex structure can be simplified and its input space

dimensions can be reduced. Adaptive and control laws can δ xm
be developed to tune all of the control gains online, θ
accommodating uncertainty in an omni-directional robot
without any learning phase [6]. T4

Over the last few years, the tracking control of four-

T3
wheeled omni-directional mobile robots [7] has become the Yω
focus of much academic interest. However, the control
structure of the CMAC cannot be utilized to solve the Xω

trajectory tracking control problem of such a robot. To

enable highly precise trajectory tracking control of the four-
wheeled omni-directional mobile robot, this work Figure 1. Structure of omni-directional mobile robot.
demonstrates a modified AFCMAC system to improve upon
the limited mathematical descriptions in the conventional B. CMAC Design
one, and to be equipped with an embedded switch robust The physical system to be controlled is assumed to have
controller to accommodate uncertainties of the multi- only one control input, and all of the state variables are
dimensional nonlinear system. assumed to be available. Accordingly, a single-output
II. STRUCTURE OF CONTROLLER CMAC is designed, with the structure that is shown in Fig. 2,
and an output that is given by
A. Omni-Directional Mobile Robot
Figure 1 shows the geometric structure of the omni- z CMAC
= F (s ) (3)
directional mobile robot. Owing to the structural symmetry
of the vehicle, its center of geometry is also the center of where F : R L → R is a nonlinear function of the CMAC
mass. The kinematic model of this robot follows, where θ input variable s = [ s1 ," , sL ]T ∈ S ⊂ R L . To mimic the
denotes the orientation of the vehicle, which is positive in the operation of the human cerebellum, the inputs (sensors) are
counterclockwise direction [7]. related to the output (response) by an association mechanism
According to the previous work [8], the kinematic model with association memory space A. Any element in A
of an omni-directional mobile robot can be described as comprises 0s and 1s, as determined by the pattern of the
follows: inputs. Vector w is the CMAC weight vector.
Mathematically, the relation (3) can be represented by a pair
ªυ1 (t ) º ª rω1 (t ) º of mappings,
«υ (t ) » « rω (t ) » ª x (t ) º
υ (t ) = « 2 » = « 2 » = P (θ (t )) «« y (t ) »» (1) G : S → A; s 6 G (s) = a(s) ∈ A (4)
«υ3 (t ) » « rω3 (t ) »

« » « » ¬«θ (t ) ¼»
¬υ4 (t ) ¼ ¬ rω4 (t ) ¼
P : A → R; a 6 P(a). (5)

ª − sin(δ + θ ) cos(δ + θ ) Lº In particular, the function P can be chosen to generate

« − cos(δ + θ ) − sin(δ + θ ) L »» output z , as follows.
P(θ (t )) = « (2) CMAC

« sin(δ + θ ) − cos(δ + θ ) L»
« » = P (a ) = a T w
¬ cos(δ + θ ) sin(δ + θ ) L¼ z CMAC
(6)

where ωi (t ), i = 1, 2,3, 4 is the angular velocity of each where a ∈ RM is the relevance vector;
wheel; r denotes the radius of each wheel, and L represents W ≡ [ w1 " w b ] ∈ R M ×b denotes the CMAC weight matrix,
the distance from the center of the platform to the center of
each wheel. and w i ∈ R M , i = 1," , b .
Figure 3 shows a possible partition of the input variables
of the CMAC, in which both s1 and s2 are divided into seven
quantified elements, so that 49 inputs states, with
m, n = −3, −2, −1, 0,1, 2,3 , are formed yielding 49 inputs
states, with m, n = −3, −2, −1, 0,1, 2,3 .[8]

840
 Weight
Association Receptive-field memory Figure 4 presents a concept map of fuzzy sets that are
memory space , A space , T space,
P :W
A → RmOutput integrated with the CMAC that presents a possible partition
G: S → A space, Y Output
Aa w1 YCMAC1 of the input variables of the CMAC. For a two-input problem,
Input
space, S
Ab w2  a fuzzy system with N fuzzy rules may be designed, each of
Ac

s1 Ba
w3
which has the form,
w4

#
(1,1)
Bb
# w5
 R (i ) : IF s1 is F1i and s2 is F2i , THEN z (fi ) is aTi w
(-2,2) Bc
sL
Ca
w6
(7)
w7
Cb w8

.. ..
. .
YCMACm
where i = 1, 2," , N , and the THEN-part is extracted from

.. ..
. .
for L=2
case Ih w26 the CMAC. Given the membership function of fuzzy set
Fki , k = 1, 2, denoted as μ F i , the following defuzzification
Ii w27
for M=27
case w1 " wm k

process is used to compute the output z FCMAC

.
Figure 2. Architecture of CMAC.

N
s2 v1a1T w + v2 aT2 w + " + vN aTN w i¦
=1
vi aTi w
z = = N (8)
v1 + v2 + " + vN
FCMAC

3.5 ¦ vi
(-3,3)(-2,3)(-1,3) (0,3) (1,3) (2,3) (3,3) i =1
2.5
(-3,2)(-2,2)(-1,2) (0,2) (1,2) (2,2) (3,2) 2
1.5 where vi = ∏ μ F i ( sk ). The preceding equation may be re-
(-3,1)(-2,1)(-1,1) (0,1) (1,1) (2,1) (3,1) k =1 k

0.5 written compactly as

0 (-3,0)(-2,0)(-1,0) (0,0) (1,0) (2,0) (3,0) s1
-0.5
(-3,-1)(-2,-1)(-1,-1)(0,-1)(1,-1)(2,-1) (3,-1)
z FCMAC
= hT Aw (9)
-1.5
(-3,-2)(-2,-2)(-1,-2)(0,-2)(1,-2)(2,-2) (3,-2)
-2.5 where
(-3,-3)(-2,-3)(-1,-3)(0,-3)(1,-3)(2,-3) (3,-3)

-3.5 -2.5 -1.5 -0.5 0 0.5 1.5 2.5 3.5

vi
h = [ h1 h2 " hN ] , hi = , A = [ a1 , a 2 ," , a N ] (10)
T T

A B N
C
¦ vi
i =1
D E F

G H I In Equation (9), matrix A (determined by the CMAC)

and vector h (determined by fuzzy rules) are typically fixed,
Figure 3. Partition of two-dimensional CMAC memory unit. but the weight vector w is adjustable herein.
D. Design of AFCMAC System
C. FCMAC Design
Based on the work of [8] and the kinematic model of the
Z omni-direction mobile robot (1), the following mathematical
N μF ( s1 ) P model of the multi-dimensional nonlinear system of the
1 1

robot is developed.
s1
-3.5 -2.5 -1.5 -0.5 0 0.5 1.5 2.5 3.5 s2
3.5 3.5 y ( n ) = f ( x ) + G ( x )u (11)
(-3,3) (-2,3) (-1,3) (0,3) (1,3) (2,3) (3,3) P
2.5 2.5
(-3,2) (-2,2) (-1,2) (0,2) (1,2) (2,2) (3,2) To merge an output feedback sliding-mode control
1.5 μ F2 ( s2 ) 1.5
(-3,1) (-2,1) (-1,1) (0,1) (1,1) (2,1) (3,1)
algorithm into the multi-dimensional nonlinear system, an
0.5 1 0.5 ideal adaptive law,
0 (-3,0) (-2,0) (-1,0) (0,0) (1,0) (2,0) (3,0) Z 0

ˆ = −γ y AT h ᧨ i = 1," , m
-0.5 -0.5
(-3,-1) (-2,-1) (-1,-1) (0,-1) (1,-1) (2,-1) (3,-1)
W i 1 i (12)
-1.5 -1.5
(-3,-2) (-2,-2) (-1,-2) (0,-2) (1,-2) (2,-2) (3,-2)
-2.5 -2.5
(-3,-3) (-2,-3) (-1,-3) (0,-3) (1,-3) (2,-3) (3,-3) N
-3.5 
-3.5 -2.5 -1.5 -0.5 0 0.5 1.5 2.5 3.5 Dˆ = γ 2 y 1 , Dˆ (0) > 0 (13)
Figure 4. Conceptual map of fuzzy sets integrated integration of fuzzy
sets with CMAC. and a control law,

841
 u = G # (x) ª¬ y (dn ) − Cx º¼ + (hT AW ˆ ⋅ sgn ( φ ) ,
ˆ )T − D (14)

are designed.
These two laws are used to construct an AFCMAC
system (Fig. 5) to accommodate the uncertainties and the
unpredictable disturbances of a multi-dimensional nonlinear
system.Next, a Lyapunov stability analysis is performed to
verify the proposition that all signals in a closed-loop system
are bounded. The Barblat lemma is then invoked to
determine that the tracking error x converges asymptotically
to zero [1, 8].
xd x u
− + y ( n ) = f ( x ) + G ( x )u
uK
x u K ≡ G # ( x) ª¬ y d(n ) − Cx º¼ + + uA

u AFCMAC
s
s = [dT x dT x ]
u AFCMAC = (hT AW
ˆ )T +
+ u AR

ˆ = −( 1φhT A)T
W Figure 6. Comprehensive physical map of experimental omni-directional
mobile robot.
 ≡ diag ª¬ φ1 ! φb º¼

u AR = − D
ˆ ⋅ sgn(φ) B. Experiment of Tracking Control
Parameters for an elliptical path are as follows.
ˆ =  2  , D
D ˆ (0) > 0

­ xd = 100 cos(θ )
°
® yd = 150sin(θ ) , 0 ≤ θ ≤ 2π (15)
Figure 5. Architecture of AFCMAC system. °θ = 0
¯ d

The initial point is ( xd (0), yd (0)) = (100, 0) and the final

III. EXPERIMENT RESULTS point is ( xd (2π ), yd (2π )) = (100, 0) . (The initial point
A. Design of Hardware of Omni-Directional Mobile Robot ( x(0), y (0)) = (0, 0) on the elliptical path, along which the
As presented in Fig. 6, The proposed controller was omni-directional mobile robot moves, is used to achieve
implemented using C/C++ code and standard programming trajectory tracking control.
techniques in the Altera Nios II embedded processor. The
FPGA chip integrated the embedded processor, RTOS, and 200
Actual path
VHDL-based IP circuits to perform the adaptive control of Desired path
150
the mobile platform.
100

50
Y(cm)

-50

-100

-150

-200
-150 -100 -50 0 50 100 150
X(cm)

Figure 7. Experiment on tracking response of AFCMAC with elliptical

trajectory.

842
Figure 7 displays the experiment on the tracking response of REFERENCES
the AFCMAC along an elliptical trajectory. The figure
verifies that the proposed AFCMAC controls an omni- [1] J. J. E. Slotine and W. Li, Applied Nonlinear Control, Prentice-Hall,
directional mobile robot with any initial point and can 1991.
perform accurate elliptical trajectory tracking. [2] Z. Y. Shen, J. Zhao, J. Xu, and X. S. Gu, “Nonlinear unknown input
observer design by LMI for lipschitz nonlinear systems,” The 8th
IV. CONCLUSION World Congress on Intelligent Control and Automation (WCICA),
pp. 3450-3454, 2010.
This work proposed a modified AFCMAC scheme that is [3] J. R. Noriega and H. Wang, “A direct adaptive neural-network control
used in the trajectory tracking control of a four-wheeled for unknown nonlinear systems and its application,” IEEE Trans.
Neural Netw., vol. 9, pp. 27-34, 1998.
omni-directional mobile robot. The proposed scheme can be
[4] W. S. Lin and C.S. Chen, “Robust adaptive sliding mode control
used to solve the tracking control problem of a class of using fuzzy modelling for a class of uncertain MIMO nonlinear
nonlinear uncertain systems. The fuzzy cerebellar model is systems,” IEE Proc. Control Theory Appl., vol. 149, pp. 193-201,
combined with both adaptive and control laws to enable the 2002.
controller gains or weights to be adjusted online. The [5] J. S. Albus, “A new approach to manipulator control: the cerebellar
FCMAC is trained to approximate the optimal weights, with model articulation controller (CMAC),” Trans. ASME, J. Dyn. Syst.
Meas. Contr., vol. 97, pp. 220-227, 1975.
the CMAC trained in 49 input states to have a high [6] T. F. Wu, P. S. Tsai, F. R. Chang, and L. S. Wang, “Adaptive fuzzy
resolution, so that the error between the input and the desired CMAC control for a class of nonlinear systems with smooth
output is small. The AFCMAC in all instances an output compensation,” IEE Proc. Control Theory Appl., vol. 153, pp. 647-
feedback sliding-mode control algorithm and ideal adaptive 657, 2006.
and control laws to reduce its tracking error to zero and to [7] H. C. Huang, “Intelligent motion control for four-wheeled
omnidirectional mobile robots using ant colony optimization,” Next
accommodate its approximation error. Wave in Robotics: Communications in Computer and Information
Finally, the proposed AFCMAC scheme is implemented Science, vol. 212, pp 94-106, 2011.
on a field-programmable gate array chip, completing the co- [8] T. F. Wu, H. C. Huang, S. S. D. Xu and N. T. Hu, “Adaptive fuzzy
design of hardware and software. The Qsys system is used to CMAC controller design for tracking of four-wheeled
present results of the simulation of the real-time trajectory omnidirectional mobile robot,” Innovation, Communication and
tracking control of the omni-directional mobile robot to Engineering-Meen, Prior & Lam (Eds), Taylor & Francis Group,
London, ISBN 978-1-138-00117-6, pp. 341-346. 2013.
demonstrate the effectiveness and the high performance of
the proposed AFCMAC.

843