Design and Simulation of Adaptive Neuro Fuzzy

Inference System (ANFIS) Controller for a Robot

Manipulator

*
Fatima Zahra Baghli , Yassine Lakhal, Larbi EI Bakkali, Ouafae Hamdoun

MODELING AND SIMULATION OF MECHANICAL SYSTEMS LABORATORY, FACULTY OF SCIENCES, UNIVERSITY ABDELMALEK

ESSAADl, BP.2121,M'HANNECH, 93002, TETUAN, MOROCCO

baghli.fatimazahra@gmail.com

Abstract- Research on the dynamic modeling and control of features of fuzzy control is that a control knowledge base is
the arms manipulators has received increased attention since the
last years due to their advantages.
available within the controller and control actions are generated
In this present work we present a new approach for a robot
by applying existing conditions or data to the knowledge base,
manipulator with two degrees of freedom based on the Adaptive
Neuro-Fuzzy Inference System (ANFIS) controller to ensure the making us of inference mechanism[4 ]-[2]. Also, the knowledge
position robot control strategy, the proposed control scheme is
based on the dynamic model derived using Lagrange-Euler
base and inference mechanism can handle no crisp, incomplete
formulation.
information; the knowledge itself will improve and evolve
Our robot manipulator ANFIS system control's simulated in
Matlab Simulink environment; the results obtained present the through learning and past experience [2].
efficiency and the robustness of the proposed control with good
performances compared with and the FIS method. Fuzzy logic control does not require a conventional model
Keywords- Robot Manipulator, FIS controller, ANFIS
Controller. of the process, whereas most conventional techniques require
either an analytical model or an experimental model. Fuzzy
I. lNTRODUCTION
logic control is particularly suitable for complex and ill-defined
The robot manipulator is a mechanical system multi
process in which analytical modeling is difficult due to the fact
articulated, in which each articulation is driven individually by
that the process is not completely known and experimental
an electric actuator is the most robot used in industry, Many
model identification is not feasible because the required inputs
efforts have been made in developing control scheme to
and output of the process may not be measurable [4].
achieve the precise tracking control of robot manipulators [1]-
[2]-[3]. The Adaptive Neuro-Fuzzy Inference System (ANFIS),
developed in the early 90s by Jang, combines the concepts of
The traditional PID controller with simple structure and
fuzzy logic and neural networks to form a hybrid intelligent
stable perfonnance is used. But it is difficult to meet the high
system that enhances the ability to automatically learn and
precision and fast response, moreover the parameters tuning of
adapt [5].
classical PID controller is so complex [11] .Therefore fuzzy
In this work after the system modeling, simulation and
algorithm is introduced. Fuzzy control is a particular type of
control robot manipulator using two articulations for motion
intelligent control, has a great potential since it is able to
using MatLab/Simulink software were carried, when the
compensate for the uncertain nonlinear dynamics using the
proposed ANFIS controlled is used to improve the articulation
programming capability of human control behavior. The main

978-1-4799-4647-1114/$3l.00 2014 IEEE

robot stability. Three types of control PID, Fuzzy logic and
ANFIS were studied; analysis and comparative studies were
Layer] Layer4
made.

The reminder paper was structured as follow: the robot

modeling is presented in second part of this paper, in the third
part of this paper ANFIS control is detailed, the results
discussion are presented in the last part of this paper and fmally
conclusion was given.

II. DYNAMIC MODEL OF ROBOT MANIPULATOR

A robot manipulator consists of a mechanical structure,
usuallay a set of rigid bodies connected in series by joints,
with the base of the robot and the end body or effector. The
dynamical equation of manipulator robot of n solids Figure.l: Structure of ANFlS
articulated between us is given by the following matrix
equation [7]-[12]: Layer 1: Every node i in the first layer employ a node
function given by:

M(q)q+C(q,q)+G(q)
ql = JlAj (x)
r =
(1)
i =1,2
(2)
Where: 0;1 = JlBj_2 (y) i = 3,4

r is the (nxl)vector of joint torque, M(q)is the (nxn)

Where JlA, and f.1Bj can adopt any fuzzy membership function
symmetric positive definite manipulator inertia matrix,

C(q,q) present the (nxl) vector of Coriolis and centrifugal (MF).

forces, G(q) is the (nxl) vector of gravitational references
Layer 2: Every node in this layer calculates the firing strength
and q, q, q are : position,velocity and acceleration of each
articulations. of a rule via multiplication:

III. ADAPTIVE NETWORK BASED Fuzzy INFERENCE i =1,2 (3)

SYSTEM POSITION CONTROLLER
Layer 3: The i-th node in this layer calculates the ratio of the
A. ANFIS Structure
i-th rule's firing strength to the sum of ail rules firing
The ANFIS is a multilayer feed forward network which
employs neural network and fuzzy logic learning algorithms to strengths:
design a plan from input to output. ANFIS has shown great
capabilities in control process [6]. A typical architecture of an
i =1,2 (4)
ANFIS is shown in Fig.l, in which a circle indicates a fixed
node, whereas a square indicates an adaptive node. For
simplicity, it was assumed that the desired logic system has
Where WI is referred to as the normalized firing strengths.
two inputs x, y and one output z. Since the proposed neuro
fuzzy model of the ANFIS similar to the first order Sugeno
Layer 4: In this layer, every node i has the following function:
fuzzy model, laws are considered as follows:
Rule 1: if (x is A I ) and (y is B I ), then (ZI = PIX + qly + 'i ). i =1,2 (5)
Rule 2: if (x is A 2 ) and (y is B2 ), then (Z2 = P2X + q2Y + r2 ).
Where w; is the output of layer 3, and {Pi, qi, r,} IS the
Where Aj and Bj are the fuzzy sets in the antecedent,
parameter set.
and PI' qj and rj are the design parameters that are determined
during the training process. As in Fig.1, the ANFIS consists of
five layers [8]:
The parameters in this layer are referred to as the consequent
II Anfis Modl Structu
parameters.

Layer 5: The single node in this layer computes the overall

output as the swnmation of all incoming signals, which is
expressed as:

(6)

The output z in Fig. 3 can be rewritten as [9, 10]:

B. Adaptive Neuro fuzzy controller

The position control of the robot arm manipulator requires
two ANFIS controllers (ANFIS1 applied to the fIrst joint and
ANFIS2 applied to the second joint).
Where two inputs and one output have been considered for the
ANFIS l, same thing for ANFIS2, the two inputs are
position), e1 (change of position error) for the fIrst controller
and el, el for
the second controller, the output is
Figure.3: The ANFlS model structure
Figures (4 and 5) show surface plot (for ANFIS l and ANFIS
2) showing relationship between inputs and output parameters
after training. The plot results from a rule base with Twenty
e1 (error fIve rules. The plateau around the origin implies a low
sensitivity towards changes in either Inputl or Input 2 near the
reference. This is an advantage if the noise sensitivity must be
low when the process is near the reference.
'I (t)

(torque) for ANFISl and '2(t) for ANFIS2.

The fIigur.4 shows the structure of control.

Arm

Manipulator

Figure.2: Arm manipulator robot ANFlS control structure Figure.4: Surface plot (ANFISl)

The strategy of the ANFIS control consists to adjust in

permanent the values of the corrector gains, the neuro fuzzy
controller developed consists of two inputs, e, and l1e,
defmes as [9].
e,(k) q; -qi
=

l1ei e(k) -e(k -1)

=

10

Where
q, is the output signal and q: is the input reference position.
The structure of ANFIS used is shown in fIgure 3,where the Figure.S: Surface plot (ANFlS2)
number of epochs was 50 for training, the memberships for the
input variables e and de is 5 and 5, respectively the number of
rules is then 25 (5*5=25) and the triangular membership
functions is used for the two input.
 IV. SIMULATION RESULTS AND DISCUS SIONS FLC2- Second Link
1.5
To show the contribution of the ANFIS and its --ref
-- Output
improvements compared to the Fuzzy logic method, a
simulation was approved on a model of two link
manipulator. This simulation was implemented in
:::l 0 - - -
Matlab/Simulink. Trajectory performance and position S-
Q'5 -0.5
error are compared in these controllers. The figures 0.
t:

follows show the behavior of the robot in pursuit of the -1

desired trajectory in three cases of control, PID, fuzzy
-1.5
logic and by ANFIS. o 2 4 6 8 10
Time [sec]
FLCI-First Link
1.5 - Figure.8.FLC 2 for the first link position
-- ref
_._._. Output
- .- ANFIS 2- Second link
1\ 2

I
g 0.5 - --- r- 1- - -- 1.5 -- ref -

5r 1 -- Output

;:;
B- ,-
0
g
'\
I I 1\ I 1\ I 1\ 1\ 1\
" 1 5r 0.5
5. -0.5 I !- - - - - f- --
1-
..:; 1 1 0
I1 o
'5
-1 1/ .., -0.5
I "
0.
IJ \, l \j V IJ
-1.5 ! .:; -1
o 2 4 6 8 10
Time[sec] -1.5

-2
Figure.6. FLC I for the first link position o 2 4 6 8 10
Tirne [sec]

ANFIS 1- First link Figure.9.ANFIS 2 for the first link position

1.5
FLCI- Position error of joint I
1.6
" '1 p\Il J\ -n
0.5 1.4

1.2
0
'5 =0
S-
-0.5 - !
O ...
0.8
-- Position error
..,
.5:::l -1 II V V \l g0)
t: 0.6
-1.5
-- ref -
:
0 0.4
output Il.

-2 I
0.2
o 2 4 6 8 10
Time [sec] 0

Figure.7.ANFIS I for the first link position -0. 2

o 2 4 6 8 10
Time [sec]

Figure.10. FLC 1 for the first link position

error
 ANFIS I-Position error of joint I PID \- Position error of joint I
1.6 2

1.4
1.5
1.2
--- Position error

..::.
0.8
2
g" -- Position error
i)
c
0.5
0.6
..g
c
g ;;; 0
;;;
.
0
.
0 0.4 0.
0-
0.2 -0.5

0
-1
0 2 4 6 8 10
o 2 4 6 8 10 Time [sec]
Time [sec]
Figure. I I. ANFIS I for the first link position Figure.14. PID I for the first link position
error error

FLe 2- Position error of joint 2

0.1 PID2- Position error of joint 2
0.4
o

" -0.1 0.2

1\ 1\ J\ " n
-- Position error
:s -0.2 A 1\ " A A " A A
::"
.g -0.3
...
g<!)
0
ft) 'V 1\1 rv V 1\1 IV 'V I\J

. ;;; c
-0.2 V U V u V V
o
0. -0.4 :
0
0. -- Position error
-0.5 -0.4

-0.6
o 2 4 6 8 10 -0.6
Time [sec] o 2 4 6 8 10
Time [sec]
Figure.12. FLC2 for the second link position
error Figure.15. PID 2 for the second link position
error

ANFIS2-Position error of joint 2

0.1 We can summaries all the obtained results in the Table. 1:
o
TABLE.I. FLC and ANFIS results
" -0.1
-- Position error
:s -0.2 Controller PID I PID FLC FLC ANFIS ANFIS
::" 2 1 2 1 2
. -0.3
Links Linkl Link Link Link Link Link
-0.4 2 I 2 1 2
Position
-0.5 error[radl 0.519 0.265 0.003726 0.000922 0.0030 0.0004
4 26 475
-0.6
o 2 4 6 8 10 We see clearly that the control performances by ANFIS are
Time [sec]
better than those in the fuzzy logic this is interpreted by the
Figure.13. ANFIS 2 for the second link position
error faster convergence of position tracking error to zero (Precision
and stability).
 v. CONCLUSION [4] M.Kevin, Passino and S. yurkovich,Fuzzy logic, Addison Wesley
longman 1998.
[5] Jang J. S. R, "Adaptive network based fuzzy inference systems",
In this present work an arm manipulator robot using two IEEE Transactions on systems man and cybernetics 1993, p. 665-685.
[6] B. Allaoua, A. Laoufi , B.Gasbaoui, and A.Aabderrahmani, "Neuro
degree of freedom was controlled using two types of controls Fuzzy DC Motor Speed Control Using Particle Swarm Optimization",
Leonardo Electronic Journal of Practices and Technologies Issue 15,
strategies , SISO control based on intelligent adaptive FLC
July-December 2009
and ANFIS, this last one present maximum control structure [7] F. Baghli, L. EI bakkali," Artificial Intelligence Application's for a
Robot Manipulator with Two Degrees of Freedom Position Control"
of our control model and give more and more efficiency for International Journal of Mechatronics, Electrical and Computer
Technology Vol. 4(11), Apr. 2014, pp. 349-368 .
the robot model with more position stability and good [8] Lin C. T., Lee C. S. G, Neural fuzzy systems: A neuro-fuzzy
synergism to intelligent systems. Upper Saddle River, Prentice-Hall,
dynamical performances with no overshoot so industrials 1996.
would take into account the efficiency of the developing [9] Constantin V. A, Fuzzy logic and neuro-fuzzy applications
explained. Englewood Cliffs, Prentice-Hall, 1995.
control model for the futures two freedom robot design [10] Kim J., Kasabov N. Hy FIS, Adaptive neuro-fuzzy inference systems
and their application to nonlinear dynamical systems. Neural Networks,
considerations. 1999.
[I I] F.Z. BaghIi, Y. Lakhal, L. EI bakkali, "Contr6le dynamique d'un bras
manipulateur it deux degrees de liberte par un contr61eur PID , I I erne
REFERENCES Congres International de Mecanique, Agadir 23 ,26 avril 2013.

[1] H.Asada & J.J.Slotine, Robot analysis and Control, New york:Wiley, [12] F.Z Baghli, 1. EI bakkali, Y. Lakhal, Abdelfatah Nasri, Brahim
1986. Gasbaoui, 'The efficiency of the inference system knowledge strate gy
for the position control of a robot manipulator with two degree of
[2] P.Sumathi,"Precise tracking control of robot manipulator using fuzzy
freedom", International Journal of Research in Engineering and
logic", DARH2005 conference, session4.1.
Technology, Volume: 02 Issue: 07 Jul-20l3.
[3] F.L. Lewis,C.T. Abdallah, and D.M. Dawson, "Control of Robot
manipulators", New York: Macmillan, 1993.