Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

Indirect Fuzzy Adaptive Control of Robotic Manipulator

Based on Sliding Mode Scheme


Feng Qiao Quanmin Zhu*, Alan Winfield, Chris Melhuish and Lifeng Zhang
Faculty of Information and Control Engineering Bristol Robotics Laboratory'
Shenyang Jianzhu University University of the West of England
9 Hunnan East Road, Hunnan New District Frenchay Campus, Coldharbour Lane
Shenyang 110168 China Bristol BS16 1QY UK
fengqiao@sjzu.edu.cn *quan.zhu@uwe.ac.uk

Abstract – An indirect fuzzy adaptive controller is designed neither an exact model of manipulator nor accurate values of
in this paper based on sliding mode scheme to tackle the tracking dynamic parameters may exist.
control issue of a 2D SCARA robot manipulator with system In the past decade, the applications of intelligent control
dynamic model uncertainties and external disturbances. The techniques (fuzzy logic control or neural-network control) to
stability of the system is ensured by the controller designed under
the motion control for robot manipulators have received
the Lyapunov’s stability theorem, and the effectiveness of the
proposed controller is verified by simulation studies with considerable attention[3]. A control system, which comprises
MatLab for the trajectory tracking control. PID control and neural network control, was presented by
Chen et al.[4] for improving the control performance of the
Index Terms – Adaptive control, sliding mode control, fuzzy system in real time. Clifton et al.[5] and Misir et al.[6] designed
systems, robot manipulation.
fuzzy–PID controllers which were applied to the position
I. INTRODUCTION control of robot manipulators. Huang and Lee[7] suggested a
stable self-organizing fuzzy controller for robot motion
Robotic manipulators are ideal candidates for material
control. This approach has a learning ability for responding to
handling operations, manufacturing and measuring devices the time-varying characteristics of a robot manipulator.
because of their capacity to pick up, move and release an However, the fuzzy rule learning scheme has a latent stability
object, to manipulate both objects and tools and their capacity
problem. Yoo and Ham[8] presented two kinds of adaptive
to explore three-dimensional space. In industrial applications, control schemes for robot manipulator via fuzzy compensator
robots usually present the features of nonlinearity and in order to confront the unpredictable uncertainties. Though
multivariable systems with inevitable modelling uncertainties the stability of the whole control system can be guaranteed,
that are required to operate with variable payloads over a some strict constrained conditions and prior system
given workspace at high speeds in a constrained environment. knowledge are required in the control process. On the other
Nonlinearities and uncertainties in robotic manipulation affect hand, Kim and Lewis[9] dealt with the application of quadratic
the precision of tracking control seriously. Many contributions optimization for motion control of robotic systems using
have been made to develop new control techniques for cerebellar model arithmetic computer neural networks. Lewis
improving the performance of the control of robotic et al. [10] developed a multilayer neural-net controller for a
manipulators. High performance specifications in these general serial-link rigid robot to guarantee the tracking
applications cannot be achieved using conventional control performance. Both system-tracking stability and error
strategies. Advanced control strategies are required that are convergence can be guaranteed in these neural-based-control
easy to understand and to design, cheap to develop, simple to systems[9,10]. However, the functional reconstructed error, the
implement, fast to respond and convenient to modify. neural tuning weights and the high-order term in Taylor series
The early stage survey paper on robotic manipulator are assumed to be known bounded functions, and some
control “Robot control systems: A survey” was published in inherent properties of robot manipulator are required in the
1987 by Moya and Seraji[1], the paper addressed the area of design process.
robot position control and gave an overview of the basic
Sliding mode control (SMC) for variable structure
problems involved. In 1991, Abdallah, et al. published another systems is a well applied technique for those systems whose
survey paper[2], discussing the robust control of rigid robots accurate mathematical models are difficult to be obtained. The
and summarised the major design methods: feedback most distinguished property of SMC is its robustness, that is,
linearisation approach, passivity approach, Lyapunov-based the closed loop systems are completely insensitive to
nonlinear approach, variable structure approach, saturation modelling uncertainties, time varying parameter fluctuations,
approach and the robust adaptive approach. and external disturbances. In spite of its wide applications, a
The issue of designing adaptive control laws for rigid pure SMC has exposed some obvious disadvantages. The first
robot manipulators that ensure asymptotic trajectory tracking one is chattering, which is highly undesirable in practical
with boundedness of all internal signals has enticed implementation, because it may excite high frequency
researchers for many years. So it has attracted great attention dynamics for the unmodelled part of a system and cause the
in this research area. Adaptive control strategies offer an system instabilities or even disasters. The other is that, in
appealing solution to robotic manipulation problems, wherein, practical implementation, it is difficult to identify an
approximation of system models as well as the model M and N can be written explicitly as
uncertain boundaries and the external disturbances. Recently  p + 2 p3 cos(q 2 ) p2 + p3 cos(q 2 )
the synthesis algorithms of modern control theory and M (q) =  1  (2)
artificial intelligence have been studied to upgrade the  P2 + p3 cos(q2 ) p2 
performance of conventional SMC. Kaynak et al. published a − q (2q + q 2 ) p3 sin( q2 )
survey paper on the fusion of computationally intelligent N (q, q ) =  2 21  (3)
methodologies and SMC[11]. Fuzzy sliding mode control,  q1 p3 sin( q2 ) 
which takes the features of both SMC and FLC to overcome where p1 , p2 and p3 are constant parameters, and
the disadvantage of chattering and enhance the robustness of
p1 = 3.1877 , p2 = 0.1168 and p3 = 0.1630 .
the controllers, is one of the examples [12,13].
In this paper, an indirect fuzzy adaptive controller based The variables are redefined as
on sliding mode scheme is designed to tackle the issue of x1 = q1 , x2 = q1 , x3 = q2 , x4 = q 2 (4)
tracking control of a SCARA robot with structure then, the matrix form of the dynamics of the SCARA type
uncertainties and parameter disturbances, and simulation manipulator is as follows
studies are made to verify the effectiveness of the proposed x = Ax + b(F ( x) + G ( x )τ + d (t ) ) (5a)
controller for tracking control of the manipulator. y = Cx (5b)
The remainder of this paper is organised as follows, in
Section , the nonlinear dynamics of a direct 2D SCARA where
x = [x1 x 2 x3 x 4 ] , y = [q1 q2 ] ,
T T
robot with model uncertainties and external disturbances is
discussed; in Section , an indirect fuzzy adaptive controller 0 1 0 0 0 0
is designed; and simulation studies are made in Section to    
0 0 0 0 1 0
verify the effectiveness of the controller developed; some A= , B= ,
conclusions are drawn in Section . 0 0 0 1 0 0
   
II. PROBLEM STATEMENT 0 0 0 0 0 1

The dynamics of the direct drive 2D SCARA robot 1 0 0 0   f ( x) 


C= , F = 1 ,
manipulator discussed in this paper is given in the following  0 0 1 0  f 2 ( x) 
matrix equation [14]:
 g ( x ) g12 ( x)  τ 
M (q)q + N (q, q ) + f c (q ) = τ (1) G =  11 , τ =  1  and
 g 21 ( x) g 22 ( x) τ 2 
q = [q1 q2 ]
T
where is the vector of joint angles,
 d (t ) 
τ = [τ 1 τ 2 ] is the torque vector applied to the joints, M (q)
T
d (t ) =  1  ;
d 2 (t )
is the inertia matrix, N (q , q ) is the vector of centripetal and
and
Coriolis forces, f c (q ) stands for Coulomb friction.
___________________________________________________________________________________________________
p p x (2 x2 + x4 ) sin( x3 ) + p3 ( p2 + p3 cos( x3 )) x22 sin( x3 )
f1 ( x) = − 2 3 4 ,
p1 p2 − p22 − p32 cos 2 ( x3 )
p3 ( p2 + p3 cos( x3 )) x 4 (2 x2 + x4 ) sin( x3 ) + p3 ( p1 + 2 p3 cos( x3 )) x22 sin( x3 )
f 2 ( x) = ,
p1 p2 − p22 − p32 cos 2 ( x3 )
p2 p2 + p3 cos( x3 )
g11 ( x) = , g12 ( x ) = − ,
p1 p2 − p22 − p32 cos 2 ( x3 ) p1 p2 − p22 − p32 cos 2 ( x3 )
p2 + p3 cos( x3 ) p1 + 2 p3 cos( x3 )
g 21 ( x) = − , g 22 ( x ) = ,
p1 p2 − p22 − p32 cos 2 ( x3 ) p1 p2 − p22 − p32 cos 2 ( x3 )
p f ( x ) − ( p2 + p3 cos( x3 )) f c 2 ( x4 ) ( p + p3 cos( x3 )) f c1 ( x 2 ) − ( p1 + 2 p3 cos( x3 )) f c 2 ( x4 )
d1 (t ) = − 2 c1 2 , and d 2 (t ) = 2 .
p1 p2 − p2 − p3 cos ( x3 )
2 2 2
p1 p2 − p22 − p32 cos 2 ( x3 )
___________________________________________________________________________________________________

It is assumed that the system states [x , x , x , x ]


1 2 3 4
T
F (x) , and the control gains G (x ) are not exactly known but
= [q1 , q1 , q2 , q 2 ] are measurable. The objective of the bounded, that is, F ( x ) ≤ F0 , and 0 < Gm G ( x) ≤ GM .
T

controller design is to derive a control law to track the


reference trajectory qr (t ) ∈ R 4 with the desired tracking III. CONTROLLER DESIGN
performance under the condition that the system functions
It is known, from the last section, that the mathematical beforehand, from (8), the control law vector u can be easily
model of the robotic manipulator to be controlled meets the determined by
following assumptions: u = −G −1 ( x)(R + F ( x ) + d − q r + H (S ) ) (11)
A 1. The system state vector x ∈ R 4 in (5) is measurable. From the Lyapunov`s candidate function selected in (7), it
A 2. The system function vector F (x) is not exactly known is easy to prove that the control law so designed can force the
and it is bounded, that is, F ( x ) ≤ F0 ( x) , where F0 ( x) is system states to track the given reference, and the tracking
a known vector, and its elements are smooth, that is, error will converge to a vicinity of zero within finite time
period. But in practical applications, it is usually difficult to
f i (x) s ( i = 1,2 ) in F (x) are smooth functions.
exactly model the manipulator in mathematical equations, or
A 3. The system gain matrix G (x ) is not exactly known, sometimes, it is impossible to get the model. In order to design
bounded by 0 < Gm ( x) ≤ G ( x) ≤ GM ( x) , where G m (x) the control law in (11), the fuzzy universal approximation
theorem is applied to approximate its parameters[16].
and G M (x) are known matrices, and G (x ) is non-
singular, and g ij (x) s ( i , j = 1,2 ) in G (x ) are smooth B. Fuzzy system and fuzzy universal approximation
A fuzzy system consists of four principle parts which are
functions.
fuzzifier (FF), fuzzy rule base (FRB), fuzzy inference engine
Here, in this section, an indirect adaptive controller is
(FIE) and defuzzifier (DF).
designed based on sliding mode scheme with fuzzy basis
FF performs a mapping of the variables
functions to approximate the unknown bounded uncertainties
x = (x1 , x 2 , ", x n ) from the crisp input domain U ⊂ R n to
T
of the manipulator to be controlled.
fuzzy domain defined in U characterising by membership
A. Design of Sliding Mode Controller
The sliding function vector is designed as function µ F : U → [0, 1] , and labelled by linguistic language,
such as “Large”, “Medium” and “Small”. The most commonly
S = [s1 , s2 ] = CE
T
(6)
used fuzzifier is “singleton fuzzifier”.
where C = [C1 , C 2 ] T
Ci = [ci ,1] ( Ci s are the Hurwitzian FRB consists of a set of linguistic rules in the form of “IF
coefficient vector), E = [e, e] is the tracking error vector a set of conditions are satisfied, THEN a set of consequences
T

are inferred”. For a fuzzy rule base with N rules, we have


whose elements are defined by the equation as e = x − qr and
R j : If x1 is A1j and x2 is A2j and " and xn is Anj Then z is B j (12)
e = x − qr .
where j = 1, 2, ", N ( N is the number of fuzzy rules), z is
The Lyapunov candidate function is selected as
1 the output of the fuzzy system, and Ai j and B j are linguistic
V = SS T (7)
2 term characterised by fuzzy membership functions µ A ( xi ) j
i
then derivation of (6) gives the following equation
and µ B (z ) , respectively. Each R j can be viewed as a fuzzy
S = [s1 , s2 ] = R + F ( x ) + G( x )u + d − q r
j
T
(8)
where implication A1j × " × Anj → B J , which is a fuzzy set in U × R
R = [R1 , R2 ] = [c1 e1 , c 2 e2 ] .
T T with µ A ×"× A → B ( x, z ) = µ A ( x1 ) ⊗ " ⊗ µ A ( xn ) ⊗ µ B ( z ) . ⊗
1
j
n
j j
1
j
n
j j

According to the reaching law method in [15], it gives is the t − norm operation, the commonly used t − norm
S = R + F ( x ) + G ( x )u + d − q r = − H (S ) (9) operation are “product” and “min”.
FIE is decision making logic which employs fuzzy rules
where
from the fuzzy rule base, to determine a mapping from the
H (S ) = Λ sgn(S ) + ΓWH (S ) (10) fuzzy sets in the input space U to the fuzzy set output space
and R.
Λ = diag [λ1 , λ2 ] Let Ax be an arbitrary fuzzy set in U , then each R j
sgn( S ) = [sgn( s1 ), sgn( s2 )] ( sgn(•) is a sign function)
T
determines a fuzzy set Ax D R j in R based on the sup-star
Γ = diag [γ 1 , γ 2 ] composition
WH ( S ) = [wh1 ( s1 ), wh 2 ( s2 )]
T
x j
[
µ A D R ( z ) = sup µ A ( x) ⊗ µ A ×"× A → B ( x, z ) =
x 1
j
n
j j ]
[ ]
x∈U
and λi , γ i are designed parameters, s i whi (s i ) > 0 and (13)
sup µ A ( x ) ⊗ µ A ( x1 ) ⊗ " ⊗ µ A ( xn ) ⊗ µ B ( z )
x 1
j
n
j j
x∈U
whi (0) = 0 ( i = 1,2 ).
DF performs a mapping from the fuzzy domain to crisp
Here, the reaching law with constant plus proportional domain. There are many techniques of defuzzification, such as
rate is applied, that is, whi ( si ) = si ( i = 1,2 ). Max Criterion (MC), Mean of Maximum (MM) and Centre of
If the system function F (x) and the control gain matrix Gravity (COG).
G (x ) are known, and G (x ) is non-singular, that is, G (x ) is If chosen COG, the crisp output of the system can be
obtained
invertible, and the external disturbance d is known
θ = [θ 1 ,θ 2 , ", θ N ]T
N N
z = ∑ µ A DR ( w j ) w j x j ∑µ Ax D R j (w j ) (14)
j =1 j =1
Theorem 1: Suppose h( x) is a continuous function on a
where w j is the point in R at which µ B (z ) achieves its j
compact set U , then for any ε ≥ 0 , there exists a fuzzy
maximum value (It is assumed that µ B = 1 ). j
system like (17), which satisfies[17]
The number of fuzzy sets, defined in the input and output sup h( x ) − z ( x ) ≤ ε (19)
universes of discourse, and the number of fuzzy rules in the x∈U

fuzzy rule base heavily influence the complexity of a fuzzy Theorem 1 states that the FBF expansions of (16) are
system, where complexity includes computational complexity, universal approximators, and the theorem is called fuzzy
i.e., the computational requirements of the fuzzy system, and universal approximation theorem.
space complexity, i.e., the storage requirements of the fuzzy C. Indirect fuzzy adaptive controller design
system. These parameters can be viewed as structure For the general form of a multi-input and multi-output
parameters of a fuzzy system. In general, the larger these nonlinear dynamic system of the manipulator in (5), it satisfies
parameters, the more complex is the fuzzy system, and the the assumptions of A1 - A3, in order to design the control law
higher the expected performance of the fuzzy system. Hence,
there is always a trade off between complexity and accuracy in in (10), the fuzzy system Fˆ ( x | θ f ) is used to approximate the
the choice of these parameters; and their choice is usually system function F ( x) , and the fuzzy system Gˆ ( x | θ G ) to
quite subjective.
The linguistic statements of the fuzzy rules are the heart approximate G ( x ) , then the control law in (10) can be written
of a fuzzy system in the sense that it is these linguistic in approximated form as
statements that contain most of the information concerning the
fuzzy system design; all other design parameters assist in the
[ −1
][
u = − Gˆ ( x | θ ) R + Fˆ ( x | θ ) + d − q + H ( S )
G (20) f r ]
effective representation and use of the information. The fuzzy and
rules usually come from two sources: human experts and [
Fˆ ( x | θ f ) = fˆ1 ( x | θ f ), ", fˆp ( x | θ f )
1
T

p
] (21)
training data.
The set of the fuzzy system above with singleton fuzzifier, = ξ ( x )θ fT
f

product inference, and Gaussian membership function consists where


of all functions of the form [
ξ f ( x) = diag ξ Tf ( x ),", ξ Tf ( x)
1 p
] (22)
N
 n  N
 n 
z ( x) = ∑  ∏ µ A ( xi ) θ j ∑  ∏ µ A ( xi ) 
[ ]
p
(15)
( n = ∑ ni )
T
j =1  i =1  j =1  i =1
i
j


i
j
θ f = θ Tf1 , ", θ Tfp ∈ R n (23)
i =1
The Gaussian membership function µ A ( xi ) is defined by
fˆi ( x | θ f ) = ξ Tf ( x )θ f i = 1, ", p
j
that is, ( ) where
[ ]
i

µ A ( x i ) = exp − (( x i − ζ i ) / σ j )
i i i

[ ]
j 2
i
j (16) ξ f ( x) = ξ f ,1 ( x)," ,ξ f ,n ( x) ∈ R n
T i
are FBFs defined as
i i i i

where ζ i , σ j are real-valued parameters, and θ j is the point


j Ni ni Ni

in R at which µ B ( y ) achieves its maximum value.


j
ξ fi ,l ( x) = ∏ µ F ( x j )
j =1
l
j
∑∏ µl =1 j =1
F jl
( x j ) ( l = 1," , ni , N i is the

 n
 n  N
 number of the rules for f i approximation). And θ f ∈ R n is
∏ µ A ( xi )  ∑∏
i

Taking   µ Ai j ( x i )  as basis
 
i
j

  j =1  i =1 
i
i =1 an adjustable vector, the membership functions µ F ( x j ) for i
j
functions and θ j as constants, z ( x ) in (15) can be viewed as
1 ≤ l ≤ m fi and 1 ≤ j ≤ k i are specified beforehand by experts’
a linear combination of the basis functions. Fuzzy basis
knowledge.
functions (FBFs) are define as below
And
 n  N  n 
g j ( x ) =  ∏
 i =1
µ Aij ( x i )  

 j =1  i =1
∑∏
µ Ai j ( x i )  , j = 1, ", N (17)

[ T
Gˆ ( x | θ G ) = Gˆ 1 ( x | θ G ), ", Gˆ p ( x | θ G ) = ξ GT ( x)θ G (24) ]
where
where µ A j ( x i ) are the Gaussian membership functions in
(16).
i
[
ξ G ( x) = diag ξGT ( x), " , ξGT ( x ) 1 p
]
[ ]
p
Then the fuzzy system (15) is equivalent to an FBF θ G = θGT1 , " , θGp
T
∈ Rn
T
( n = ∑ ni )
expression: i =1
N
that is,
z ( x) = ∑ g j (x )θ j = ξ θ T
(18)
j =1 Gˆ i ( x | θ G ) = ξ GT ( x)θ G i i
( i = 1, ", p ) (25)
where
ξ = [ξ 1 , ξ 2 ,", ξ N ]T = [g1 (x ), g 2 (x ), ", g N ( x )]T
and
[ ]T
where ξ G ( x ) = ξ G ,1 ( x ),",ξ G ,n ( x ) ∈ R n are FBFs defined
i The initial condition of the system in this case is selected
as to [x1 x 2 x3 x 4 ]T = [0.4 1.2 0.2 0.1]T .
i i i i

Ng ni Ng

as ξ Gi,l ( x) = ∏ µG ( x j )
j =1
l
j
∑∏ µ
l =1 j =1
Glj
( x j ) ( l = 1,", ni , N g is The Hurwitzian coefficient in (6) is designed as c1 = 8
and c 2 = 12 . Five fuzzy sets are designed for each of the crisp
the number of fuzzy rules). And θ G ∈ R n ×n is an adjustable i i
i
variables x1 , x2 , x 3 and x 4 on (x1 × x2 × x3 × x4 ) =
matrix, the membership functions µ G ( x j ) for 1 ≤ l ≤ mGi and
i
j ([0,1] × [− 1,1]× [0,1]× [− 0.5,0.5]) , and each fuzzy set is labelled
1 ≤ j ≤ N g are specified beforehand by experts’ knowledge. as NB for Negative Big, NS for Negative Small, ZE for ZEro,
PS for Positive Small, and PB for Positive Big. The
Theorem 2: For the control system (5), if the assumptions A1- membership functions of the fuzzy sets for x are selected in
A3 are satisfied and the control law vector is designed by (20)
evenly distributed Gaussian shapes. H (S ) in (10) is designed
with (21) and (22), and the parameter vectors θ f and θ G are
as
adjusted by the following adaptation laws
s  s 
 1  H (S ) = diag [0 0]sgn  1  + diag [20 8] 1  . (32)
Pr oj  ξ f ( x )S − θf , ϕ f  = 0 (26)  s2   s2 
 γf 
  The adaptation rates γ f 1 , γ f 2 and γ G for f 1 (t ) ,
and
f 2 (t ) and g ij (t ) ( i, j = 1,2) are chosen as γ f 1 = 8 , γ f 2 = 6
 1 
Pr oj  ξ G ( x )S − θG , ϕ G  = 0 (27) and γ G = 6 , respectively.
 γG  The simulation results are shown in Fig. 2.
( Pr oj is the vector projection operation, and γ f and γ G are
B. Case study 2
the adaptation rates respectively, positive constants)
The reference tracking trajectory functions for Case 2 are
then the closed loop system signals will be bounded and the
assumed to be as follows
tracking error vector defined in (6) will be convergent to zero
asymptotically. qr1 (t ) = 0.5 × (cos(2 × t ) + 1) (33)
Theorem 2 can be proved in the similar way as that in q r1 (t ) = sin(2 × t ) (34)
[16]. qr 2 (t ) = 0.5 × (sin(3 × t ) + 1), 0 ≤ t ≤ Ts (35)
IV. SIMULATION STUDIES q r 2 (t ) = 1.5 × cos(3 × t ), 0 < t ≤ Ts (36)
Simulation studies are made with MatLab to verify the And the reference trajectories are shown in Fig.3.
effectiveness of the proposed indirect adaptive controller And the initial condition of the system in this case is
selected as to [x1 x 2 x3 x 4 ] = [0.8 0 0 1.5] . Others
T T
developed for the direct drive 2D SCARA robot manipulator
described in Section Ⅱ. are assumed to be the same as those in Case 1.
As the system functions f1 ( x) and f 2 ( x ) and the control The simulation results are shown in Fig.4.
gains g11 ( x) , g12 ( x) , g 21 ( x) and g 22 ( x ) in (5) are not Within the two case studies, the reference tracking
trajectories are selected as sine shapes and ramp shape. The
exactly known, the initial conditions of θ f 1 (0) , θ f 2 (0) , simulation results show the desired tracking performance of
θ G11 (0) , θ G12 (0) , θ G 21 (0) and θG 22 (0) randomly selected the controller designed.
within [− 1 1] . The sampling time instant in the simulation V. CONCLUSIONS
studies is set to ∆t = 0.01 second, and the simulation period is An indirect fuzzy adaptive law based on sliding mode
selected as Ts = 10 seconds[18]. scheme is proposed for the multi-input and multi-output
tracking control problem of a 2D SCARA robot manipulator
A. Case Study 1
with system uncertainties, the proposed algorithm takes the
In Case 1, the reference tracking trajectory functions for
advantages of sliding mode control, fuzzy system and adaptive
the simulation are assumed to be as follows
control with reaching law method and fuzzy universal
qr1 (t ) = 0.5 × (sin(3 × t ) + 1), 0 ≤ t ≤ Ts (28)
approximation, and these is no need to know much about the
qr1 (t ) = 1.5 × cos(3 × t ), 0 < t ≤ Ts (29) structure and bounds of the parameters of systems as in the
 2 design of conventional sliding mode control. The stability of
 t , if 0 ≤ t ≤ Ts / 2 the control system is proved in terms of the Lyapunov second
qr 2 (t ) =  Ts (30)
 1, stability theorem. The two simulation case studies are made in
if Ts / 2 < t ≤ Ts
this paper to show the effectiveness of the proposed hybrid
 2 / Ts , if 0 ≤ t ≤ Ts / 2 control algorithm.
qr 2 (t ) =  (31)
 0, if Ts / 2 < t ≤ Ts REFERENCES
The reference trajectories are shown in Fig. 1. [1] M. M. Moya and H. Seraji, “Robot control systems: A survey”, Robotics,
Vol. 3, Issues 3-4, September-December, 1987, pp. 329-351.
[2] C. Abdallah, D. M. Dawson, P. Dorato and M. Jamshidi, “Survey of the
Robust Control of Robots”, IEEE Control Systems Magazine, Vol. 11,
No. 2, February, 1991, pp. 24-30.
[3] R. J. Wai, “Tracking control based on neural network strategy for robot
manipulator”, Neurocomputing, 51, 2003, pp. 425-445.
[4] P. C. Y. Chen, J. K. Mills, G. Vukovich, “Neural network learning and
generalization for performance improvement of industrial robots”,
Canadian Conference on Electrical and Computer Engineering, 1996,
pp. 566.
[5] C. Clifton, A. Homaifar and M. Bikdash, “Design of generalized Sugeno
controllers by approximating hybrid fuzzy–PID controllers”, IEEE
International Conference on Fuzzy Systems, 1996, pp. 1906-1911.
[6] D. Misir, H. A. Malki and G. Chen, “Graphical stability analysis for a
fuzzy PID controlled robot arm model”, IEEE International Conference
on Fuzzy Systems, 1998, pp. 451-455.
[7] S. J. Huang and J. S. Lee, “A stable self-organizing fuzzy controller for
robotic motion control”, IEEE Transactions on Industrial Electronics,
Vol. 47, Issue 2, 2000, pp. 421-428. Fig. 2 Tracking trajectory and tracking error of Case 1
[8] B. K. Yoo and W. C. Ham, “Adaptive control of robot manipulator
using fuzzy compensator”, IEEE Transactions on Fuzzy Systems, Vol. 8,
Issue 2, 2000, pp. 186-199.
[9] Y. H. Kim and F. L. Lewis, “Optimal design of CMAC neural-network
controller for robot manipulators”, IEEE Transactions on Systems Man
and Cybernetics, Vol. 30, Issue 1, 2000, pp. 22-31.
[10] F. L. Lewis, A. Yesildirek and K. Liu, “Multilayer neural-net robot
controller with guaranteed tracking performance”, IEEE Transactions on
Neural Networks, Vol. 7, Issue 2, 1996, pp. 388-399.
[11] O. Kaynak, K. Erbatur, and M. Ertugrul, The Fusion of Computationally
Intelligent Methodologies and Sliding-Mode Control: A Survey, IEEE
Transactions on Industrial Electronics, Vol. 48, No. 1, 2001, pp. 4-17.
[12] J. Park, J. Kim, and D. Park, LMI-based design of stabilizing fuzzy
controllers for nonlinear systems described by Takagi-Sugeno fuzzy
model, Fuzzy Sets and Systems, Vol. 122, 2001, pp. 73-82.
[13] W. Chang, J. B. Park, Y. H. Joo, and G. Chen, Design of robust fuzzy- Fig 3 Reference trajectories for Case 2
model-based controller with sliding mode control for SISO nonlinear
systems, Fuzzy Sets and Systems, Vo. 125, 2002, pp. 1-22.
[14] K. Erbatur, O. Kaynak, A. Sabanovic and I. Rudas, “Fuzzy adaptive
sliding mode control of a direct drive robot”, Robotic Autonomous
Systems, Vol. 19, No. 2, 1996, pp. 215-227.
[15] W. B. Gao, and J. Hung, “Variable structure control of nonlinear
systems: a new approach”, IEEE Transactions on Industrial Electronics,
Vol. 40, Issue 1, February, 1993, pp. 45-55.
[16] F. Qiao, Q. M. Zhu, A. Winfield, C. Melhuish, “Adaptive sliding mode
control for MIMO nonlinear systems based on fuzzy logic scheme”,
International Journal of Automation and Computing, Vol.1, 2004, pp.
51-62.
[17] L.X. Wang, Adaptive fuzzy systems and control, design and stability
analysis, PTR Prentice-Hall, Englewood Cliffs, NJ, 1994.
[18] F. Qiao, “Fuzzy Sliding Mode Control of Complex Dynamic Systems
and its Applications”, PhD Thesis, University of the West of England, Fig.4 Tracking trajectory and tracking error of Case 2
Bristol, UK, 2005.

Fig. 1 Reference trajectories for Case 1

You might also like