Ijcsi 9 4 3 275 285

Genetic Algorithm Based PID tuning for Controlling
Paraplegic Humanoid Walking Movement

Hashim Ali
[1]
, IkramUllah
[2]
, M.Irfan
[3],
Mohsin Shahzad
4],
M. Aftab
[5]

[1][2][3][4][5]
Comsats Institute of IT,Abbottabad.

Abstract: Genetic Algorithm (GA) is a
very useful tool to search and op timize
many engineering and scientific
problems. In this paper, a real tim e
enhanced biomedical model of
humanoid structure is developed in MSC
visual Nastran to assist the paraplegic
patient. The complexity of the model is
driven by the needs that the model
parameters must be estim ated for an
eventual individual with disability. After
the development of humanoid structure
an inverse model is designed to estim ate
the joint torques. The re ference
trajectories of the humanoid m odel are
obtained from MSC visual Nastran.The
controllers are des igned in M atlab/
Simulink for f our joints which are
manually tuned simultaneously to obtain
the results. Afterwards, GA is used to
tune the P ID controllers to find the
optimal solutions which are com pared
with manually tuned PIDs. The re sults
are shown and hence it is proved that
GA has given a better optim ized control
system.
Keywords: Genetic Algorithm, PID,
Inverted Pendulum, Optimization,
Walking Movement.

1. Introduction

Paraplegia is a disease caused by the
spinal cord injury (S CI) paralyzing
lower body of the patient. Patients
suffering from this disease are known as
paraplegics. Paraplegics experience

immobility or lack of activity in lo wer
body part. Recently, a lot of work has
been done in this area. Due to the large
varieties of humanoid structure it is not
possible to apply a general control
method to all humanoid m odels.
Intelligent control techniques like Neural
Networks, Fuzzy control, Genetic
Algorithm (GA) etc., have been used to
control humanoid models due to their
evolutionary learning, faster
convergence and good tolerance of
uncertainty and imprecision [1]. As these
intelligent techniques are independent of
model in optimization, and only depends
upon the feedback from the system to
improve performance. As the precise
dynamic model of hum anoid is not
available due to the complex mechanical
structure and high degr ee of freedom, so
the evolutionary techniques are useful in
humanoid control.
Humanoid walking is a hot issue in
many research publications but this issue
is still unsolved or addressed, because of
its dependency on certain other issues
like 3-D walking, impact of effects
between humanoid robot and ground
plane [2]. Pratt et al. have developed
RoboKnee [3] to support the knee
motion during walking or clim bing up
stairs. Johnson et al have developed a
support system using pneumatic muscles
by compressed gas not for a healthy
person but for the paralyzed, am putee
and spastic patient [4]. One of the aim s
of this work was to solve problems about
weight, power, endurance and cost of the
IJCSI International Journal of Computer Science Issues, Vol. 9, Issue 4, No 3, July 2012
ISSN (Online): 1694-0814
www.IJCSI.org 275
Copyright (c) 2012 International Journal of Computer Science Issues. All Rights Reserved.

actuators. Many authors have
investigated control of muscle force by
either computer sim ulation or
experiment. Optimal control of an
antagonistic muscle pair has been
discussed in [5-6] and implem entations
of closed-loop control of a single muscle
have been also reported [7-10].
However, since their works were
restricted to single -input single-output
systems, it is very difficult to understand
how to control multiple muscles in order
to regulate a multi joint body movement.
In humanoid research field, m any
researchers have used GA as an
optimization tool [11-13]. A GA based
trajectory generation m ethod for a
prismatic joint humanoid robot has been
presented by Capi et al [11]. Hasegawa
et al [12] has presen ted a hierarchical
evolutionary algorithm to generate a
natural walking m otion on the slope
surface to minimize total energy. Cheng
and Lin [13] have used GA to
investigate a control design to search a
suitable control and param eters of
trajectory based on the robustness
measure developed from the liberalized
Poincare map. However, all of them
have not considered 3-D humanoid
walking and restricted to the motion to a
single plane independently.
In this paper, a sim pler technique and
not like passive dynam ic walk but an
electrically controlled system on a
humanoid model is used. The
development of the real tim e humanoid
model is being done in MSC visual
Nastran 4-D to obtain the reference
trajectories. Matlab/ Simulink is used to
implement the controllers f or the
disabled lower pa rts of the body to
initiate walking m ovement and the
controllers are tuned both manually and
GA based to f ollow the ref erence
trajectories for four joints obtained from
MSC visual Nastran 4-D.
This paper is organized as follows.
Section 2 presents the developm ent of
the humanoid model, humanoid walking
gait and its phases. Section 3 covers the
design of controller structure. A brief
review of GA and the proposed GA
based tuned PID controller is described
in section 4. The results are discussed in
section 5.Finally the conclusion is given
in section 7.

2. Humanoid Model

2.1 Humanoid Structure

The development of an accu rate
humanoid model is extremely important.
Designing an accu rate controller to
perform walking m ovement is not
possible until and unless the re is an
accurate model. Because of the high
degree of freedom and complex
mechanical structure it is hard to
precisely model the dyna mics of
humanoid. To build an accura te model a
computer aided design tool is required
that can simulate the designed model in
real time. This m odel is created in
AutoCAD and is im ported to MSC
visual Nastran software. It can be easily
connected with Matlab/ Simulink so that
a suitable controller can be designed for
making the m odel to m ove. The
complete development of the hum anoid
model includes all the body segm ents
and measurements of t heir length and
weights. The main focus of this paper is
on the left hip, left knee, right hip, and
right knee while the rest of the body
movement is ignored. Revolute motors
are being us ed on these specified joints
to obtain the reference trajecto ries in
terms of instantaneous angles for a
www.IJCSI.org 276

normal human walking as shown in
figure 1.

Fig.1 Humanoid Model Developed.

2.2 Walking Gait

A unique feature of locom otion analysis
is that unpredictable hum an behavior
must be taken into account when the
control system is designed [14]. The
main difficulty lies in balancing the
body, reducing the contact force between
the feet and step s and avo iding
slipping of the f eet in the support
phase. To balance the body and to avoid
foot slipping, the weight acceptance
(moving centre of m ass forward) is
essential. The above developed model is
moved in MSC visual Nastran to obtain
the instantaneous positions of left hip,
left knee, right hip and right knee in
terms of angles f or a tim e interval of
three seconds as shown in figure 2.

Fig.2 Walking gait with reference
trajectories.
2.3 Phases of Walking

The first phase is the initiation phase.
During this phase the forward
momentum is generated. The forward
momentum generation is n ecessary in
order to tra nsfer weight from the still
position to a m oving position. The
beginning of this phase corresponds to
fast changes in the anterior-posterior
component of the ground reaction f orce.
The initial phase f or a normal walking
movement can be considered as a
ballistic movement. Table 1 shows the
left and right foot com plete 2-steps
cycle.
S.No Phase Activity Cycle
1
Initial
Phase
Trunk movement with
still legs
10%
2
First
Step/Left
foot
Forward Acceleration
10%-
40%
3
Back to
initial
position
Deceleration/Balancing
Body
40%-
50%
4
Second
Step/Right
foot
Forward Acceleration
50%-
80%
5
Final
phase
Back to still position
80%-
100%

Table 1: Phases of the walking movement
3. Controller Structure Design

To design a controller that will con trol
the movement of the hum anoid model
according to the obtained reference
trajectories. The PID controllers are used
to track the reference trajectories of the
revolute motors used for each left hip,
left knee, right hip, and right knee. In
this design, the voluntary upper body
effort has been taken as a disturbance.

3.1 Multivariable PID Control System

To achieve satisfactory perform ance,
multivariable control methods m ust be
used. This is a s trong motivation to
derive effective yet sim ple methods for
tuning PID controllers used with in a
multivariable controller. Consider the
case of 44-square system.

3.1.1 Matrix Vector Expression
The system can be modeled as:
( ) y f u u =
(1)
www.IJCSI.org 277

where output and control vectors
satisfying
4
, y ue9
and 4 4-square
matrix of system transfer fun ction
4 4
( ) f u

e9

The error can be written as:
l h l h l h
l k l k l k
r h r h r h
r k r k r k
e r y

( ( (
( ( (
( ( (
( ( (

=
(2)
where error, reference and output vectors
satisfying
4
, , e r y e9

and the controller of the system will be :
( )
l h l h
l k l k
r h r h
r k r k
c
u G s e

( (
( (
( (
( (

=
(3)
where control and error v ectors
satisfying
4
, e u e9
and 4-sqaure matrix
of controller transfer function satisf ying
4 4
( )
c
G s

e9

3.1.2 Element wise Matrix Vector
Definition
The system equation
( )
l h l h
l k l k
r h r h
r k r k
y f u u

( (
( (
( (
( (

=
is given the
element wise description.
( )
y u
l h l h
y u
l k l k
y u
r h r h
y u
r k r k
f u

( (
( (
( (
(
=

( (
( (

(4)
where output and con trol components
are:
, , , , , , , ( )
l h l k r h r k l h l k r h r k
y y y y u u u u s

e9
,
The error equatio n
l h l h l h
l k l k l k
r h r h r h
r k r k r k
e r y

( ( (
( ( (
( ( (
( ( (

= is given by
the element wise description.
e r y
l h l h l h
e r y
l k l k l k
e r y
r h r h r h
e r y
r k r k r k

( ( (
( ( (
( ( (
=
( ( (
( ( (

(5)
where error, reference and output vector
elements satisfying:
4
,
, , , , , , , , , ,
l h l k r h r k l h l k r h r k l h l k r h r k
e e e e r r r r y y y y

e9

Finally the contr oller equation
( )
l h l h
l k l k
r h r h
r k r k
c
u G s e

( (
( (
( (
( (

= is given the
element wise description.
-
-
-
-
0 0 0
0 0 0
0 0 0
0 0 0
u e
l h l h l h
u e
l k l k l k
u e
r h r h r h
u e
r k r k r k
g
g
g
g

( ( (
( ( (
( ( (
=
( ( (
( ( (

(6)
where the error and control vector
elements satisfy:
4
, , , , , , ,
l h l k r h r k l h l k r h r k
e e e e u u u u

e9

Figure 3 shows the control system for
multivariable model.

Fig. 3 Multivariable PID Control System

The integration of Matlab/ Simulink
with MSC visual N astran makes it
www.IJCSI.org 278

possible for Simulink to send signals to
MSC visual Nastran and to receive
signals from it in real tim e. In this way
the two software environm ents work
together. Therefore, a control system in
Matlab/ Simulink is e ssential to e nable
the joints angular displacement, angular
velocity, angular acceleration and
torques to track the d esired trajectories
and the hum anoid model to perform
walking in MSC visual Nastran. As there
are various com ponents of the system
which are dynamically strongly coupled,
slight change in any joint causes
disturbance and va riation in the
trajectories of all other segments. Hence,
it is no t easy to con struct a precise
mathematical model that describes the
dynamic behavior of the humanoid
model. A closed loop control system has
been designed to ach ieve the desired
trajectory.

4. Problem Formulation

In order to move the humanoid model to
walk under the control of PID controllers
the reference trajectories are needed,
which should be followed for the
balance motion of the model. First of all
the reference trajecto ries are generated
in MSC v isual Nastran for norm al
human walking and then PID controllers
are used to move the model accord ing to
the reference trajectories. To tun e the
PID controllers so that the model follows
the reference trajectories is a very time
consuming process and also it cannot
guarantee good walking performance.
In the proposed method, a multi-criteria
GA is used to search o ptimal value for
the parameters of the PID contr oller
which will guarante e to f ollow the
reference trajectories. So the objectiv e
function can be defined in terms of
errors and PIDs ar e tuned using GA to
minimize the error.

4.1 Objective Function

To design the four GA based PID
controllers so tha t it will f ollow the
reference trajectories given by the visual
Nastran 4-D. The objective function can
be defined as:
3 3 3 3
0 0 0 0
(7)
l h l k r h r k
LeftHip LeftKnee RightHip RightKnee
J w edt w edt w edt w edt

| | | | | | | |
= + + +
| | | |
\ . \ . \ . \ .
} } } }

l h l h l h
l k l k l k
r h r h r h
r k r k r k
e r y
( ( (
( ( (
( ( (
( ( (
( ( (
( ( (

=
(8)
where e is the error, r is the reference
trajectory and y is the current output of
the plant and
1
l h l k r h r k
w w w w

= = = =
.

4.2 Optimization Problem

In this study, m inimization of the
objective function J is used. The
constraints are the co efficients of the
PID controller.
Minimize J
Subjected to:
min max
, , ,
min max
, , ,
min max
, , ,

P i P i P i
I i I i I i
D i D i D i
K K K
K K K
K K K
s s
s s
s s

where
[1, 2,3, 4] i e
. The proposed
design is optim ized through GA to
search the optimal controller parameters.
Table 2 gives the details of the
parameters with their bounds to be
optimized by GA.

www.IJCSI.org 279

S.No Parameters Min Max
1 K
P1
0 50
2 K
I1
0 50
3 K
D1
0 50
4 K
P2
0 50
5 K
I2
0 50
6 K
D2
0 50
7 K
P3
0 50
8 K
I3
0 50
9 K
D3
0 50
10 K
P4
0 50
11 K
I4
0 50
12 K
D4
0 50
Table 2: The parameter bounds for
optimization
In Table 3 the param eter values of GA
based tuned PIDs are given.
S.No Parameter Value
1
Maximum
Generations 100
2
Population
Size
20
3
Crossover
Probability

0.6
4
Mutation
Probability
0.1
Table 3: GA parameters

5. Humanoid Walking using GA
based Control System

The genetic algorithm is an optimization
algorithm which sim ulates the natural
evolution process to search the optimal
solution of the optimization problem. It
depends entirely on responses from its
environment and evolution operators
(i.e., reproduction, crossover and
mutation) to arrive at the best solution.
By starting at several independent points
and searching in parallel, the algorithm
avoids local minima and converging to
sub optimal solutions.
In this pape r, GA is used to search the
parameters of the PID contro llers in
terms of objective function minimization
as shown in figure 4.

Fig.4 Flowchart of Design process

A closed loop control system is designed
for humanoid model movement as
shown in figure 5.

Fig.5 Interface of MSC visual Nastran with
Matlab/ Simulink

In this f igure it is shown that the
humanoid model is moved in MSC
visual Nastran and its m ovement is
controlled by closed loop control system
in Matlab/ Simulink. The parameters of
the PID controllers of closed loop
control system are tuned using GA
implemented in Matlab.

6. Results and Simulations

The normal humanoid walk has been
carried out in MSC visual Nastran for a
www.IJCSI.org 280

time interval of 3 seconds to obtain the
reference trajectories. Figure 6 shows the
initial and final positions of the
humanoid model in MSC visual Nastran.

Fig.6 Initial position and final position of
humanoid model in MSC visual Nastran

A closed loop control system is applied
to move the m odel and the PID
controller parameters are tuned manually
and using GA. The angular
displacement, angular velocity, and
angular acceleration corresponding to
the reference trajectories are g iven in
figures 7, 8 and 9 respectively.

Fig.7 Angular Displacement

Fig.8 Angular Velocity

Fig.9 Angular Acceleration

In figure 10 the torques plots are given
corresponding to the reference
trajectories for left hip, left knee, right
hip and right knee respectively.

Fig.10 Torques
www.IJCSI.org 281

In this figure, the torque plots of the
knees are continuous because of the fast
variations at each point of hum anoid
walk while the to rque plots of the hips
are discrete because of the slow
variations at each point.

6.1 Optimization Results

Both manually and GA based tuned PID
controllers parameter values are given in
Table 4.
S.No Parameters
Value

Manually
tuned
GA
tuned
1 K
P1
42 21
2 K
I1
0 39
3 K
D1
0.7 31
4 K
P2
15 11.5
5 K
I2
1 4.6
6 K
D2
0.6 5
7 K
P3
23 5.5
8 K
I3
0 21
9 K
D3
2 30.5
10 K
P4
33 6
11 K
I4
1 1.6
12 K
D4
0.99 30

Table 4: Parameters values of manually
tuned and GA based tuned PID
controllers
Figures 11-14 show the tracked
reference trajectories by manually tuned
PID based control system for left hip,
left knee, right hip and right knee
respectively.

Fig.11 Reference and achieved left knee
trajectory

Fig.12 Reference and achieved left hip
trajectory

Fig.13 Reference and achieved right
knee orientation trajectory

Fig.14 Reference and achieved trunk
right hip trajectory
While the plots of reference trajectories
following by GA ba sed tuned PID
controllers of the closed loop control
system are given in figures 15-18 for left
hip, left knee, right hip and right knee
respectively.
www.IJCSI.org 282

Fig.15 Reference and achieved left hip
trajectory

Fig.16 Reference and achieved left knee
trajectory

Fig.17 Reference and achieved right
knee trajectory

Fig.18 Reference and achieved right hip
trajectory
The static and dyna mic performance of
both manually and GA based tuned PID
controllers can easily be d epicted
through the IAE and IT AE as described
from equations below:
3
2
0
IAE : e dt
}

3
2
0
ITAE : ( ) t e dt
}

Figure 19 shows the IAE value of both
GA based and m anually tuned PID
controllers.

Fig.19 Comparison between the PIDs
tuned manually and PIDs tuned by GA
for IAE
It is observed that the IAE value of GA
based tuned are less than that of
manually tuned PIDs for all f our joints.
Figure 20 shows the ITAE value of both
GA based and m anually tuned PID
controllers.
www.IJCSI.org 283

Fig.20 Comparison between the PIDs
tuned manually and PIDs tuned by GA
for ITAE
It is observed that the ITAE value of
GA based tuned are less than that of
manually tuned PIDs for all f our joints.
The combined IAE and ITAE values for
both GA based and m anually tuned of
four PID controllers are given in figure
21.

Figure 21: Comparison between the
combined PIDs tuned manually and
PIDs tuned by GA for IAE and ITAE

7. Conclusion

In this paper, hum anoid model is
developed in MSC visual Nastran.The
reference trajectories have been obtained
from normal humanoid walk in MSC
visual Nastran. A closed loop control
system with PID controllers is designed
in Matlab/ Simulink to contro l the
humanoid movement in MSC visual
Nastran to fol low the reference
trajectories by tuni ng the PIDs. The
PIDs are tuned both manually and using
GA and their respective performances
are shown through the results. It is
observed that GA has tuned the P IDs
better than that of manually tuned by
minimizing the error.

References
[1] Katic D, Vukobratovic M, Survey of
Intelligent Control Techniques f or
Humanoid Robots, Journal of Intelligent
and Robotic Systems, 37:117-141, 2003.
[2] Hemami H, Zhen g Y.F, I mpact
Effects of Biped Contact with the
Environment, IEEE Tans. System s.
Man. and Cybernetics, SMC-14(3):
437-443, May/June 1984.
[3] Benjamin T. Krupp, Collins Steven
H, Christopher J. Morse and Pratt
.Jerry E, The RoboKnee: An
Exoskeleton for Enhancing Strength
and Endurance during W alking,
Proceedings of the 2004 IEEE
International Conference on
Robotics and Autom ation (ICRA
04), pp.2430-2435, 2004.
[4] David C. Johns on, Daniel W.
Repperger and Thompson Greg,
Development of a mobility assist for
the paralyzed, amputee, and spastic
patient, Proceedings of the IEEE
Southern Biomedical Engineering
Conference, pp. 67-70, 1996.
[5] Buys.J.D and H. Hatze, Energy-
optimal controls in the mammalian
neuromuscular system, Biological
Cybernetics., vol. 27, pp. 9-20, 1977.
[6] Oguztoreli M.N. and Stein R. B,
Optimal control of antagonistic
muscles, Biol. Cyhern, vol. 48, pp.
91-99, 1983.
[7] Bemotas.L.A ,Crago P. E, and
Chizeck H.J, Adaptive control of
electrically stimulated muscle, IEEE
Trans. Biomedical Engineering, vol.
BME-34, pp. 140-147, 1987.
[8] Crago P. E, Mortim er J. T, and
Peckham P. H, Closed-loop control
www.IJCSI.org 284

of force during electrical stim ulation
of muscle, lEEE Transactions on
Biomedical Engineering, vol. BME-
27, pp. 306-312, 690-761, 1980.
[9] Chizeck H. J and Peterson D.K,
Linear quadratic control of a loaded
agonist antagonist muscle pair, IEEE
Trans. Biomedical Engineering, vol.
BME-34, pp. 790-796, 1987
[10] Chizeck H.J, Crago P.E and
Wilhere G.F, Design and
evaluation of a digital closed-loop
controller for the reg ulation of
muscle force by recruitm ent
modulation, IEEE Trans
Biomedical Engineering, vol.
BME-32, pp. 668-676, 1985.
[11] Barolli L, Capi G, Kaneko S,
Mitobe K and Nasu Y, Opt imal
trajectory generation for a
prismatic joint biped robot using
genetic algorithms, Robotics and
Automation Systems, 38: 119-128,
2002.
[12] Arakawa T, Fukuda T and
Hasegawa Y, Trajectory generation
for biped locom otion robot,
Mechatronics, 10: 67-89, 2000.
[13] Cheng M.Y, and Lin S, Dynam ic
biped robot locom otion on less
structured surfaces, Robotica,
18:163-170, 2000.
[14] Kaphle Manindra, Sim ulations of
human movements through
temporal discretization and
optimization, KTH Mechanics,
2007.

Hashim Ali has been in
research since 2006.He
has received
masters in computer
engineering(2009)
with distinction. He has
also done bachelors in computer science
(2006). Currently he is peer reviewer of
IEEE and member international journal
applied engineers.

www.IJCSI.org 285

Ijcsi 9 4 3 275 285

Uploaded by

Copyright:

Available Formats

Ijcsi 9 4 3 275 285

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Ijcsi 9 4 3 275 285

Uploaded by

Copyright:

Available Formats

Genetic Algorithm Based PID tuning for Controlling

Paraplegic Humanoid Walking Movement

You might also like