2010 IEEE/ASME International Conference on

Advanced Intelligent Mechatronics

Montréal, Canada, July 6-9, 2010

Fuzzy Pre-compensated Fuzzy Self-Tuning Fuzzy PID Controller of

3 DOF Planar Robot Manipulators

A. F. Amer, E. A. Sallam, Member, IEEE, and W. M. Elawady, Computers and Automatic

Control
containing both structured uncertainties (parametric) and
Abstract—Control of an industrial robot includes unstructured uncertainties (un-modeled dynamics). Variable
nonlinearities, uncertainties and external perturbations that
should be considered in the design of control laws. structure control (VSC) is one of the robust control strategies
Proportional-integral-derivative (PID)-type fuzzy controller is to compensate these uncertainties in robotic dynamics. In
a well-known conventional motion control strategy for
manipulators which ensures global asymptotic stability. To these robust control design approaches [8, 9], a fixed control
enhance the PID-type fuzzy controller performance for the
control of rigid planar robot manipulators, in this paper, a law based on a priori bound of uncertainty is designed to
fuzzy pre-compensation of a fuzzy self tuning fuzzy PID compensate the effects of system uncertainties.
controller is proposed. The proposed control scheme consists of
a fuzzy logic-based pre-compensator followed by a fuzzy self A fuzzy logic controller (FLC) makes control decisions
tuning fuzzy PID controller. In the fuzzy self tuning fuzzy PID
controller, a supervisory hierarchical fuzzy controller (SHFC) by its well-known fuzzy IF– THEN rules. FLCs can be
is used for tuning the input scaling factors of the fuzzy PID classified into two major categories: the Mamdani type FLC
controller according to the actual tracking position error and
the actual tracking velocity error. Numerical simulations using that uses fuzzy numbers to make decisions [7] and a Takagi–
the dynamic model of a three DOF planar rigid robot
manipulator with uncertainties show the effectiveness of the Sugeno (TS) type FLC that generates control actions by
approach in set point tracking problems. Our results show that linear functions of the input variables. In the early years,
the proposed controller has superior performance compared to
a conventional fuzzy PID controller. most FLCs were designed by trial and error. Since the
complexity of a FLC will increase exponentially when it is
Keywords— Fuzzy PID, robots, supervisory fuzzy controller,
uncertainties used to control complex systems, it is tedious to design and
tune FLCs manually for most industrial problems like
I. INTRODUCTION
robotics systems. This is why the conventional nonlinear
I N the recent decades, the motion control of robot
manipulators has received a great of attention. Motion design method [15] was adopted in the fuzzy control area,

control is needed to make each joint track a desired such as fuzzy sliding control, fuzzy gain scheduling [24],

trajectory as close as possible. Many control algorithm such and adaptive fuzzy control [18], in order to alleviate

as computer torque method [12], [19], optimal control [1-2], difficulties in constructing the fuzzy rule base.

adaptive control [13], [23], variable structure control (VSC) Analytical calculations show that a two-input FLC

[4], [14], neural networks (NNs) [22] and fuzzy system [3], employing proportional error signal and velocity error signal

[6], [8], [20], have been proposed to deal with this robotic is a nonlinear proportional-integral (PI) or proportional-

control problem. In [12], [19], a computer torque control is derivative (PD) controller [16], [11]. Due to the popularity

developed on the basis of the feedback linearization. of PID controllers in industrial applications, most of the

However, these designs are possible only the dynamics of development of fuzzy controllers revolves around fuzzy PID

the robotic dynamic are well known. controllers in the past decade [21]. PI- and PD-type Fuzzy

Generally, uncertainties may not be known in practical controller have gained prominence for almost two decades

robotic systems such as changing payload, nonlinear friction, now because of their simple structure, ease of

unknown disturbance, and the high frequency part of the implementation and inherent robustness [5], [17].

dynamics. Therefore, it is necessary to consider these effects Another possible implementation of a PID-type FC requires

1
978-1-4244-8030-2/10/$26.00 ©2010 IEEE 599
the development of a popular PD- type Fuzzy controller in • • •
e(t ) = q d − q that will be fed to fuzzy controller while
parallel with a conventional integral (I) controller. Presently
the research interest in this field is focused on the integral error ³ e(t ) dt will be use as conventional integral
development of adaptation policies which can adapt these
action. Figure 1 shows the structure of Fuzzy PD+I
component fuzzy controllers in a suitable and simple
controller.
fashion, and furthermore can achieve reasonably accurate
d q
and satisfactory performance. An excellent example of Kd Fuzzy PD +
u
dt KU Robot
Control +
applying self-organized fuzzy systems to auto tune the gains q+d e Kp
−
of a classical PID controller online and its successful
application in controlling an MIMO robot arm can be found Ki
³ e(t) dt
in [10].
In this paper utility and effectiveness of soft computing Fig. 1 Block diagram of fuzzy PD+I controller
•
approaches for the control of three degree of freedom planar The linguistic descriptions of inputs ( e(t ) , e(t ) ) and
robot arm with structured and unstructured uncertainties is
output u membership functions are NL (negative large),
presented. In this approach, pre-compensation of a hybrid
NM (negative medium), NS (negative small), Z (zero), PS
fuzzy PD+I controller is proposed. The control scheme
(positive small), PM (positive medium) and PL (positive
consists of a fuzzy logic-based pre-compensator followed by
large). The fuzzy membership functions for inputs and
fuzzy PD+I control. Moreover, a fuzzy supervisory
output are shown in Figures 2 and 3. The control surface of
controller is used to supervisor a conventional proportional
the output u is shown in figure 4.
and derivative actions such that the conventional gains are NL
1
NM NS Z PS PM PL

adapted on-line through fuzzy reasoning. 0.8

Degree of membership

Simulations are performed on MATLAB to show the

0.6

effectiveness of the proposed adaptive fuzzy algorithm.

0.4

Comparative analysis of simulated performance of

0.2

conventional controllers is also presented to validate the

0

controller design. The remainder of this paper is organized -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
Rate of change of error

as follows: section II presents the characteristics of a fuzzy

PD+I controller, section III presents a fuzzy pre- Fig. 2 Membership functions of inputs ( e(t ) , e(t ) )
•

compensated fuzzy PD+I controller, section IV presents the

NL NM NS Z PS PM PL
characteristics of a fuzzy self-tuning fuzzy pre-compensated 1

fuzzy PD+I controller, section V presents the simulation 0.8

Degree of membership

results, and section VI summarizes the conclusions and 0.6

contributions of the works. 0.4

0.2

II. FUZZY PD + I CONTROLLER 0

-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1

In this section fuzzy PD+I had been applied to the output

dynamic model of three DOF planar robot arm. Three input Fig. 3 Membership functions of output u
had been used, position error e(t ) = q d − q , change of error

2
600
 ^
modified desired position q d (t ) is simply the sum of the

external desired position q d (t ) and γ (t ) . The fuzzy sets of

0.5

the inputs and the output are shown in figure 6 and figure 7.
output

Figure 8 shows the fuzzy control surface of the output γ (t ) .

-0.5

The fuzzy pre-compensator is described by:

1
0.5 1 •
0 0.5 ~ ~ ~
-0.5
-0.5
0
q (t ) = q d (t ) − q (t ) , γ (t ) = F [q(t ), q(t )]
Rate of error -1 -1
error
^

Fig. 4 Control surface of output u q d (t ) = q d (t ) + γ (t )

The input scaling factors are set as: The input scaling factors associated with the position error
K p = diag{0.9, 0.9, 0.9} and K d = diag {0.7, 0.7, 0.7} . The ~
•
~

output scaling factor KU is set to 5000. The conventional q(t ) and the velocity error q(t ) are set to K e = 0.5
and K de = 0.2 , respectively. The output scaling factor K u is
integral gain K i is set to 1.
set to 2 .
III. FUZZY PRE-COMPENSATED FUZZY PD+I NL
1
NM NS Z PS PM PL

CONTROLLER
0.8

In this work, a fuzzy self tuning fuzzy pre-compensation

Degree of membership
0.6

fuzzy logic-based PD+I control is attempted. The fuzzy pre- 0.4

compensator modifies the desired position by introducing a 0.2

fuzzy logic controller in front of the fuzzy PD+I controller. 0

-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1

Figure 5 shows the overall block diagram of the fuzzy pre- error

•
~ ~
compensated fuzzy PD+I controller. Fig. 6 Membership functions of q (t ) and q(t )
^
Modified qd NL NM NS Z PS PM PL
1

qd +
Ke Fuzzy + 0.8
+ −
Precom Ku
Degree of membership

-
d K de pensato 0.6

dt 0.4

0.2

d Kd q
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1

e Fuzzy + u output
dt KU Robot
PD
Control
+ Fig. 7 Membership functions of output γ (t )
+ Kp
−

Ki
³ e(t ) dt 0.5
output

Fig. 5 Block diagram of fuzzy pre-compensated fuzzy PD+I controller

-0.5

•
~ ~ 1

In our description, we think of q (t ) and q(t ) as inputs, 0.5

0 0.5
1

0
-0.5
-0.5
and γ (t ) as the output. The term γ (t ) represents a Rate of error -1 -1
error

compensation or correction term, so that the compensated or Fig. 8 Fuzzy control surface of output γ (t )

3
601
 IV. FUZZY-SELF TUNING FUZZY
PRECOMPENSATED FUZZY PD+I CONTROLLER
To obtain the better controller performance over a wide 1.1

operating range, in this paper, a typical fuzzy self-tuning 1

fuzzy pre-compensated fuzzy PD+I controller structure is

Kp
implemented as shown in Figure 9. 0.9

0.8
^
Modified qd
1
0.5 1
0 0.5

qd Ke Fuzzy -0.5 -0.5

0
+ -1 -1
+ −
Precom Ku Rate of error error
-
d K de pensato
dt Fig. 10 Fuzzy control surfaces of Kp

Supervisory
Fuzzy Control
0.8

d Kd q 0.7

e Fuzzy + u
dt KU Robot

Kd
PD
+ 0.6
Control
+ Kp 0.5
−

³
1
Ki e(t ) dt 0.5
0.5
1
0
0
-0.5 -0.5
Rate of error -1 -1
error

Fig. 9 Block scheme of fuzzy self-tuning fuzzy pre-compensated fuzzy

PD+I control Fig. 11 Fuzzy control surfaces of Kd

Specifically, a fuzzy inference system is used to NB NS Z PS PB

adaptively tune the input scaling factors of the direct fuzzy 1

controller K p and K d depending on the current operating 0.8

Degree of membership

conditions of the controlled system. The goal of adaptive 0.6

fuzzy control is to maintain performance in terms of

stability, tracking error, or other specifications, despite 0.4

parametric uncertainty, external disturbances, un-modeled 0.2

dynamics, or other uncertainties present in the system.
Figures 10 and 11 show the fuzzy control surfaces of K p 0

-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1

and K d respectively. These surfaces have been used to error

adaptively tune K p and K d on line. The gaussian shape •

Fig. 12 Membership function of e(t ) and e(t )
membership functions with 50% of overlapping for the input
and output fuzzy variables are considering as shown in
Z PS PM PB PVB
figure 12, figure 13, and figure 14. The physical domain of 1

•
the inputs ( e(t ) , e(t ) ) is in the range {-1, 1} and that of the 0.8
Degree of membership

outputs ˱( K p , K d ) is in the range {0.7, 1.2} and {0.4, 1.0} 0.6

respectively, selected based on trial and error approach. 0.4

0.2
V. SIMULATION RESULTS
0
The proposed controller was tested for the control of a 3
0.7 0.75 0.8 0.85 0.9 0.95 1 1.05 1.1 1.15 1.2
Kp
DOF rigid three–link manipulator shown in figure 15.
Fig. 13 Membership functions of Kp

4
602
 1.2
Z PS PM PB PVB Fuzzy PD + I
1
Fuzzy precompensated Fuzzy PD + I
1
Fuzzy self-tuning fuzzy precompensated Fuzzy PD + I

0.8

Position error of joint 1 (rad.)

0.8

Degree of membership
0.6 0.6

0.4 0.4

0.2
0.2

0
0

-0.2
0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Kd Time t (sec.)

Fig. 17 Tracking error of joint 1

Fig. 14 Membership functions of Kd 1.5
Desired position
Fuzzy PD+I
Fuzzy precompensated fuzzy PD+I
Fuzzy self-tuning fuzzy precompensated fuzzy PD+I

Position of joint 2 (rad.)

1

τ3 θ3
0.5

τ2 0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
θ2 Time t (sec.)

τ1 Fig. 18 Position of joint 2

θ1 1.2
Fuzzy PD+I
Fuzzy precompensated fuzzy PD+I
1
Fuzzy self-tuning fuzzy precompensated fuzzy PD+I

Fig. 15 Rigid three-link robot manipulator

Position error of joint 2 (rad.)
0.8

For the case of direct fuzzy controller with fixed input 0.6

scaling factors, K p = diag {0.9, 0.9, 0.9} and K d = diag {0.7, 0.7, 0.7}. 0.4

0.2

Figures 16, 18, and 20 show the set point tracking for joints
0

1, 2, and 3, respectively. Figures 17, 19, and 21 show the -0.2

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time t (sec.)
joint position tracking error profile for joints 1, 2, and 3
Fig. 19 Tracking error of joint 2
respectively. Table 1 reiterates this claim with the help of 1.5
Desired position

steady state error and setting time calculated for each of the Fuzzy PD+I
Fuzzy precompensated fuzzy PD+I
Fuzzy self-tuning fuzzy precompensated fuzzy PD+I

competing controllers for each joint.

Position of joint 3 (rad.)

1.5
Desired position
Fuzzy PD + I
Fuzzy precompensated Fuzzy PD + I 0.5
Fuzzy self-tuning fuzzy precompensated Fuzzy PD + I
Position of joint 1 (rad.)

0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time t (sec.)
0.5
Fig. 20 Position of joint 3

0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time t (sec.)

Fig. 16 Position of joint 1

5
603
 1.2 [2] A. Green, J.Z. Sasiadek, Fuzzy and optimal control of a two-link
Fuzzy PD+I flexible manipulator, in: Proceedings of the IEEE/ASME
Fuzzy precompensated fuzzy PD+I
1
Fuzzy self-tuning fuzzy precompensated fuzzy PD+I International Conference on Advanced Intelligent Mechatronics ,
pp. 1169–1174, 2001.
Position error of joint 3 (rad.)
0.8
[3] B.K. Yoo, W.C. Ham, Adaptive control of robot manipulator
0.6 using fuzzy compensator, IEEE Trans. Fuzzy Syst. 8 ,186–198,
2000.
0.4
[4] C.Y. Su, T.P. Leung, A sliding mode controller with bound
0.2 estimation for robot manipulators, IEEE Trans. Robotics
Automat. 9, 208–214, 1993.
0
[5] D. Driankov , H. Hellendorn, Reinfrank M, An introduction to
-0.2
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
fuzzy control, New York: Springer-Verlag, 1993.
Time t (sec.)
[6] E.A.M. Cruz, A.S. Morris, Fuzzy-GA-based trajectory planner
Fig. 21 Tracking error of joint 3 for robot manipulators sharing a common workspace, IEEE
Trans. Robot. 22, 613–624, 2006.
Table 1 steady state error and setting time [7] E. H. Mamdani and S. Assilian, An experiment with in linguistic
synthesis with a fuzzy logic controller, Int. J. Man-Machine
Settling time Stud., vol. 7, pp. 1–13, 1975.
Steady state error (rad.)
Algorithm (sec.) [8] F.C. Sun, Z.Q. Sun, G. Feng, An adaptive fuzzy controller based
on sliding mode for robot manipulators, IEEE Trans. Syst. Man
J1 J2 J3 J1 J2 J3 Cybern. 29 ,661–667, 1999.
Convention
al fuzzy 0.002 0.005 0.005 4.4 3.6 4.2
[9] G. K. I. Mann, B. G. Hu, and R. G. Gosine, “Analysis of direct
PD+I action fuzzy PID controller structures,” IEEE Trans. Fuzzy Syst.,
vol. 29, pp. 371–388, June 1999.
Fuzzy
precompens
[10] H. B. Kazemian, The SOF-PID controller for the control of a
ated fuzzy
0.001 0.0025 0.0004 2.2 2.3 2.3 MIMO robot arm. IEEE Trans Fuzzy Syst 10: 523–532, 2002.
PD+I [11] H. Ying, “A nonlinear fuzzy controller with linear control rules
is the sum of a global two-dimensional multilevel relay and a
self-tuning local nonlinear proportional-integral controller,” Automatica,
fuzzy
vol. 29, pp. 499–505, 1993.
precompens 0.009 0.0001 0.0003 1.2 1.2 1.2
ated fuzzy [12] J.J. Craig, Introduction to Robotics, second ed., Addison-Wisley,
PD+I Reading, MA, 1989.
[13] J.E. Slotine, W. Li, Composite adaptive control of robot
manipulator, Automatica 25 (1989) 509–519.
From the simulation results, it is concluded that our [14] J.E. Slotine, W. Li, Applied Nonlinear Control, Prentice-Hall,
Englewood Cliffs (NJ), 1991.
proposed fuzzy pre-compensated fuzzy self-tuning fuzzy [15] J. J. E. Slotine and W. P. Li, Applied Nonlinear Control. Upper
Saddle River, NJ: Prentice-Hall, 1991.
PD+I controller showed superior performance for each of [16] K. J. Astrom and T. Hagglund, PID Controllers: Theory,
the three joints. Hence it can be inferred that our proposed Design, and Tuning. Research Triangle Park, NC: Instrum. Soc.
Amer., 1995.
controller could outperform another existing popular [17] K. M. Passino, S. Yurkovich (1998) Fuzzy control. Menlo Park,
CA: Addison-Wesley-Longman
variation of fuzzy controller with gain tuning provision. [18] L. X. Wang, Adaptive Fuzzy Systems and Control: Design and
Stability Analysis. Upper Saddle River, NJ: Prentice-Hall, 1994.
[19] M.W. Spong, M. Vidyasagar, Robot Dynamics and Control,
VI. CONCLUSIONS Wiley, New York, 1989.
The present paper has described the development of a [20] M.A. Llama, R. Kelly, V. Santibanez, Stable computed-torque
control of robot manipulators via fuzzy self-tuning, IEEE Trans.
fuzzy pre-compensated fuzzy self-tuning fuzzy PD+I Syst. Man Cybern. 30, 143–150, 2000.
controller with input gain tuning in which it has been [21] M. J. Er and Y. L. Sun, Hybrid fuzzy proportional-integral plus
successfully applied for the motion control of robot conventional derivative control of linear and nonlinear systems,
manipulators. It could also prove a success in a challenging IEEE Trans. Ind. Electron., vol. 48, pp. 1109–1117, Dec. 2001.
domain of robot manipulators where the dynamics for each [22] O. Barambones, V. Etxebarria, Robust neural control for robotic
link was expressed by complex, nonlinear, time-varying, manipulators, Automatica 38, 235–242, 2002.
coupled differential equations. An interesting research topic [23] R. Ortega, M.W. Spong, Adaptive motion control of rigid
robots: a tutorial, Automatica 25, 877–888, 1989.
is concerned with whether this controller can perform, as
[24] R. Palm, D. Driankov, and H. Hellendoorn, Model Based Fuzzy
well as fuzzy robust controllers and adaptive fuzzy sliding Control. New york: Springer-Verlag, 1997.
mode controllers in their relevant domains, without actually
embodying the significant complexities of these robust and
sliding mode controllers.

REFERENCES
[1] A. Green, J.Z. Sasiadek, Dynamics and trajectory tracking
control of a two-link robot manipulator, J. Vibr. Control 10,
1415 –1440, 2004.
6
604