Wheelchair Control and Navigation Based on

Kinematic Model and Iris Movement

Razvan Solea, Adrian Filipescu, Adriana Filipescu Jr.

Department of Automation and Electrical Engineering,
“Dunarea de Jos” University of Galati, Romania

Silviu Filipescu
Eugenia Minca Department of Electrical Drives
Department of Automation, Computer Science and “Politehnica” University of Bucharest, Romania
Electrical Engineering
“Valahia” University of Targoviste, Romania

Abstract — This paper deals control and navigation of the data from encoders that is robust against localization errors.
wheelchair for elderly and disabled based on kinematic model This controller is a path-following sliding-mode controller, for
iris motion and image processing. Cirrus Power Wheelchair was a car-like robot, claiming robustness to localization and
modelled as an wheeled mobile robot (WMR) with two driving curvature estimation errors. Also, this paper proposed a
and two free wheels (2DW/2FW). An input/output model of the navigation system based on iris and eye movements, allowing
wheel control system consisting of servo-amplifier, DC motor and obstacle avoidance.
gear reducer have been experimentally indentified. To control
each driving wheel system, PI controller has been used. Pole The rest of the paper is organised as follows: in Section II,
placement method was used for tuning PI controller. A software nonholonomic wheelchair kinematic model is laid out; driving
solution was proposed for real-time computing of the wheelchair wheel control system is presented in Section III; Identification
position based on data acquisition from whell encoders. A of driving wheel DC motor model is shown in Section IV;
navigation system, based on iris movement and image processing, tuning PI controller parameters is done in Section V; a solution
was real-time implemented in MATLAB. This navigation system for data acquisition from encoders, is elaborated in Section VI;
identifies the user's eye movement in three directions: forward, wheelchair navigation based on iris movement and image
left and right. Stop of the wheelchair is done by eye closing. Using processing together a graphic user interface designed in
a LabView platform, a graphic user interface has been designed LabView are presented in Section VII,; some final remarks can
and implemented, allowing user to control wheelchair
be found in Section VIII.
movements.
II. NONHOLONOMIC WHEELCHAIR KINEMATIC MODEL
Keywords—wheelchair; control; navigation; kinematic model;
iris movement. Fig 1 shows a geometric model of a wheelchair that defines the
main variables necessary to obtain kinematic model [1].
I. INTRODUCTION
The progress in mobility of the wheeled mobile robots
(WMR) made possible its application in various and useful
fields, like the use of intelligent powered wheelchairs to assist
handicapped and elderly people. Most of the research work on
robotic wheelchairs (and differential drive vehicles in general),
incorporate classic solutions for low level control and
concentrate design efforts on high level control (path following
and trajectory tracking). The literature addressing the problem
of motion under nonholonomic constraints using the kinematic
model of a mobile robot is vast. On the other hand, little has
been written about the integration problem of the
nonholonomic kinematics controller and the dynamics of the
mobile robot. Furthermore, there is a small amount of works on
robustness and control in presence of uncertainties in the
dynamical model of such systems. Several results on the
control of nonholonomic systems with uncertainties have been
obtained. These results concern the back-stepping technique
applying to the adaptive control of nonholonomic systems with Fig. 1. Wheelchair modelled as wheeled mobile robot with two driving
unknown parameters. This paper proposes a path following wheels and two free wheels (2DW/2FW)
controller based on WMR’s kinematic model and acquisition

978-1-4673-7338-8/15/$31.00 2015
c IEEE 78
 This WMR has two diametrically opposed drive wheels A(q )S (q ) = 0 (8)
(radius r) and free-wheeling castors. Both drive wheels are
actuate and sensed, while the castors are neither actuated nor where the constant c = r (2b ) . The kinematic model, is given
sensed. Thus, the castor wheels are not considered in getting of by
kinematic and dynamic models. P0 defines the origin of the
q = S ( q )v (t ) (9)
robot coordinate system with coordinates (xo , yo ) . Pc is the
center of mass of the robot with coordinates (xc , yc ) and is with
placed in the X -axis at a distance d of Po . Pl is a virtual v = [v R [
v L ]T , ν = θ R θ L ]T (10)
reference point attached to the platform with coordinates
(xl , yl ) and is placed in the X -axis at a distance La of Pc . Concerning the robotic system in study, Δ distribution is
2b is the length of the axis between the wheels of the mobile involutive and the system is controllable.
robot. φ is the angle between x-axis of {W } and the X -axis III. DRIVING WHEEL DC MOTOR MODELLING
of {R} . θ and θ are the angular velocities of the right and
r l To control a DC motor must have the mathematical model
left wheels around the Y-axis. of the motor [2]. The mathematical model must be able to
determine a relationship between the angular velocity of the
The equilibrium of the wheelchair is maintained by two rotor and its applied voltage.
front free wheels and other two rear small wheels, whose effect
it will ignore. Thus, To get the input/output model of wheel DC motor, having
as input the armature voltage and as output angular speed, we
q = [xc , yc , φ, θr , θl ] (1)
can write the steady-state
(1) mechanical and electrical variables:
dθ
ω(t ) =
denotes the configuration of the system, i.e. the five dt (11)
generalized coordinates (n = 5) . In the kinematic model it is
dθ
suppose that in each contact exist a pure rolling motion. eb (t ) = K b
Assuming that the velocity of Po must be in the direction of
dt (12)
the symmetry (X-axis) and the wheels do not slip, the following dia
constraints set (m = 3) , with respect to Pc , is obtained [8]: ea (t ) = Ra ia (t ) + La + eb (t )
dt (13)
y c cos φ − xc sin φ − φ d = 0 (2)
(2) Tm (t ) = Kt ia (t ) (14)
xc cos φ + y c sin φ + bφ − rθ r = 0 (3) dω
Tm (t ) = J
xc cos φ + y c sin φ − bφ − rθ l = 0 (4)
dt (15)
These constraints can be rewritten in the matrix form: where: Ra is armature resistance, La is inductance armature
(5)
Tm is active torque, ia is current through the armature circuit;
A( q ) q = 0
ea is input electromotive voltage, eb is counter electromotive
with voltage, ω is angular speed, J is moment of inertia of the
ª − sin φ cos φ − d 0 0º motor shaft and K b , K t are motor constants.
A(q ) = ««− cos φ − sin φ − b r 0»» (6) Setting U a (s ) = Ea (s ) as control input, then it can express,
«¬− cos φ − sin φ b 0 r »¼ in Laplace, the transfer from the control input to the angular
speed output as follow
Considering the mobile robot kinematics S (q ) is defined Ω(s ) = H (s )U a (s )
by: (16)
After some calculus manipulations, the transfer function
ªc(b cos φ − d sin φ) c(b cos φ + d sin φ)º (16) becomes
« »
«c(b sin φ + d cos φ) c(b sin φ − d cos φ)» Km
S (q ) = « c −c » (7) H (s ) =
« » JTa J
1 0 s2 + s +1
« » K a Kt Kb K a Kt Kb
« » (17)
¬ 0 1 ¼
where: K m = 1 Kb , Ta = La Ra , K a = 1 Ra .
that satisfies the equation

2015 IEEE 7th International Conference on CIS & RAM 79

Because the s 2 coefficient is small ( Ta is small) compared to
the coefficient of s , then we can neglect the term containing
s 2 . Therefore, the transfer function (17) becomes in
approximate form
Km
H (s ) =
τs + 1 (18)
where: τ = J (K a K t Kb ) .
To control DC motors we need mathematical model of the
whole driving wheel system consisting of servo amplifier, DC
motor and speed reducer [3].
For controlling brush DC motors were used servo amplifier
50A8, shown in Fig.2. It can receive data from acquisition
board and, based on the received signal, the servo-amplifier
will send a PWM control signal to DC motor, signal having a
duty cycle dynamically modifiable. For data acquisition and
control signal sending to the DC motors has been a NI-6024
acquisition card, also shown in Fig. 2. It has 16 analogue
inputs, two analogue outputs 8 digital inputs and 8 digital
Fig. 3. Equipped Cirrus Power Wheelchair
outputs.
The communication between the motherboard and data IV. DRIVING WHEEL DC MOTOR MODEL IDENTIFICATION
acquisition card is a PCI slot. This device connects to PC the To identify the transfer functions of the two wheel control
servo-amplifies, DC motors and the encoders (Fig. 2). PC systems MATLAB’s SYSTEM IDENTIFICATION toolbox,
processes signals and data necessary for odometer system [4]. This toolbox allows identifying parameters of transfer
implementation. Cirrus Power Wheelchair fully equipped with function (18) the mathematical model of systems from
PC, data acquisition card, servo-amplifiers, DC motors, input/output acquisition data from each driving wheel control
encoders and batteries is shown in Fig. 3. system.
To obtain the parameters of transfer function for each
diving wheel control system (servo amplifier, DC motor,
angular speed reducer) we have to acquire output data (angular
speed) applying a voltage signal to the input.
The data acquired from the two sensors that measuring the
angular speed of the two drive wheels are showin Fig. 3 and
Fig 4.
Angular speed for left motor
12
input
output
10
Angular speed [rad/sec]

-2
0 5 10 15 20 25
Time [sec]

Fig. 4. Left DC motor angular speed to an input variable step voltage

Fig. 2. Control and navigation hardware (voltage applied to servo amplifier)

80 2015 IEEE 7th International Conference on CIS & RAM

 Angular speed for right motor Angular speed for left motor with PI controller
12 6
input desired
real
10 output 5
Angular speed [rad/sec]

Angular speed [rad/sec]

8 4

6 3

4 2

2 1

0 0

-2 -1
0 5 10 15 20 25 0 5 10 15 20 25 30
Time [sec] Time [sec]

Fig. 5. Right DC motor angular speed to a variable step voltage input(voltage Fig. 6. Closed loop with PI controller, left DC motor angular speed response
applied to servo amplifier) to a variable step reference input

Due to parameter identification, were obtained the Angular speed for right motor with PI controller
6
following transfer functions corresponding to each wheel (left
desired
and right) control sytem real
5

Angular speed [rad/sec]

2 .6
HL = 4
0.28s + 1 (19)
3
2.7
HR =
0.3s + 1 (20)
2

1
V. TUNING PI CONTROLLER PARAMETERS
Since the transfer functions of the control systems for both 0
drive wheels, are first-order, then it will use a PI controller. By
using pole placement method, ([3], [4], [7]) allows to compute -1
0 5 10 15 20 25 30
parameters of both PI controllers, K P and Ki . Time [sec]
Pole placement method revealed the following values:
Fig. 7. Closed loop with PI controller, left DC motor angular speed response
to a variable step reference input
K PL = 0.6 , K i L = 2 , K PR = 0.5 , K i R = 2.1 (21)
θ step k current step − θ step k previous step
The experimental results using PI controllers embedded in ω=
WINDOWS REAL-TIME TARGET in MATLAB are shown
t (21)
in the Figs. 6 and 7.
360D π
VI. SOFTWARE SOLUTION FOR DATA ACQUISITION FROM θ step = ⋅
number of pulses per rotation 180D
ENCODERS (22)
To implement a software solution we use PCI-6024 card
and dedicated MATLAB’s toolboxes, SIMULINK and REAL- where: θ step is the angle of DC motor shaft, k current step is
TIME WINDOWS TARGET, that allow real-time work. the number of pulses to the current step, k previous step is the
The resolution of encoders is 10,000 pulses per revolution number of pulses to the previous step, t is sample time. The
corresponding to a sample time of 0.0001sec. Since the sample number of pulses per rotation represents the total number of
time is very small, an external signal is involved to increase pulses achieved a complete rotation of the encoder.
sample time up to 0.1sec for data acquisition and computing.
In Fig. 8 are shown acquisition data computed by the above
The angular speed of each wheel can be computed using formulas.
the following relationships:

2015 IEEE 7th International Conference on CIS & RAM 81

Fig. 8. Real-time data acquisition for left and right DC motor angular speed

VII. WHEELCHAIR NAVIGATION BASED ON IRIS MOVEMENT

AND IMAGE PROCESSING
The navigation system consists of: image processing Fig. 9. Phases for iris and its center identification
system, graphic user interface and wheelchair control. Imaging
system is to define the controls for steering.
The whole set of information in the image processing [5],
[6], [7] includes eye detection, direction finding and validating
it by averaging 10 frames. Eye position generates a control
signal that can be shaped string forward, left, right, forward
direction, left or right or when you want to stop actuators
STOP command will be generated by undetected iris eye
closure. These commands are sent to the graphical user
interface made in LabView programming environment. The
steps for identifying the iris and its center are shown in Figure
9.
To identify the direction based on the position of iris
circum-center has to measure the center position vs image Fig. 10. Method for direction identification
position along the axis OX (0-320 pixels). When the center of
the circle is below 100, the direction will be left; when is over
200, direction is right, and for the center it will take long for
intermediate values. The direction signals will be sent to the
graphic user interface as strings: FORWARD - for forward
direction, LEFT – for left direction, RIGHT – for right
direction and STOP – for stopping of the wheelchair.
Wheelchair stops when is undetected iris, i.e. eye is closed.
We have defined two safety belts on OY (0 ÷ 240 pixels),
an upper in the range (180 ÷ 240) and lower in the range (0 ÷
40). These strips are designed to filter out false circles that may
result from involuntary eye movements. Figure 10 describes
the implementation of the solution direction finding by the eye
movements of the user.
The graphic user interface, created with a LabVIEW
platform, is shown in Fig 11. It was highlighted with red
arrows for inactive direction and green for the direction chosen
by the user.
When evaluating experimental results and their accuracy,
were taken into account a number of factors such as processing
power of the computer that equips the wheelchair disturbances
in iris identification etc.

Fig. 11. Graphic user interface for Cirrus Power Wheelchair’s control and
navigation

82 2015 IEEE 7th International Conference on CIS & RAM

 ACKNOWLEDGMENT
The work of Adrian Filipescu, Eugenia Minca and Silviu
Filipescu was supported by the Romanian Executive Unit of
Funding Higher Education, Research, Development and
Innovation (UEFISCDI), project number PN-II-ID-PCE-2011-
3-0641, project title: Advanced control of reversible
manufacturing systems of assembling and disassembling using
wheeled mobile robots equipped with robotic manipulators.
The work of Razvan Solea and Adriana Filipescu was
supported by the Romanian Executive Unit of Funding Higher
Education, Research, Development and Innovation
(UEFISCDI), project number PN-II-PT-PCCA-2013-4-0686,
project title: Prototypes of autonomous robotic systems for
medical/social assistance and servicing of manufacturing
processes in metallurgy, ceramics, glass and automotive.
The work of Razvan Solea was partially supported by the
Sectoral Operational Programme Human Resources
Development 2007-2013 of the Ministry of European Funds
through the Financial Agreement POSDRU/159/1.5/S/132397.
The work of Silviu Filipescu has been partially funded by
the Sectoral Operational Programme Human Resources
Development 2007-2013 of the Ministry of European Funds
through the Financial Agreement POSDRU/159/1.5/S/134398.

Fig. 12. Wheelchair navigation based on rela-time isris and eye movement
REFERENCES
VIII. CONCLUSIONS [1] Dhaouadi R. and A. Abu Hatab, Dynamic Modelling of Differential-
Control and navigation of the wheelchair based on Drive Mobile Robots using Lagrange and Newton-Euler Methodologies:
kinematic model and image processing of iris movement are A Unified Framework, Advances in Robotics & Automation, Vol. 2(2),
pp. 1-7, 2013.
presented. The wheelchair is modelled as an WMR
[2] Chen X., J. G. Chase, P. Wolm, I. Anstis, J. Oldridge, W. Hanbury-
autonomous system, with 2DW/2FW. Driving wheel control Webber, R. Elliot, W. Pettigrew, System Identification and Modelling
system are modelled and parameter identified by using of Front Wheel Drive Electric Wheelchairs, 17th International
MATLAB’s System identification toolbox. Pole placement Federation of Automatic Control World Congress (IFAC 2008), 6 pages,
method was used for tuning PI controller parameters. A 2008.
software solution was proposed for real-time computing of the [3] Braga, R. A. M., M. Petry, L. P. Reis, A. P. Moreira, IntellWheels:
wheelchair position based on data acquisition from the wheel Modular development platform for intelligent wheelchairs, Journal of
Rehabilitation Research & Development, Vol. 48(9), pp. 1061-1076,
encoders. A navigation system, based on iris movement and 2011.
image processing, was implemented by using MATLAB’s
[4] Ljung, L., System Identification Toolbox - For Use with MATLAB, The
REAL-TIME WINDOWS TARGET and IMAGE MathWorks, Inc., 1995.
ACQUISITION toolboxes. This navigation system identifies [5] B. Siciliano, L. Sciavicco, and G. Oriolo, Robotics: Modelling,
the user's eye movement in three directions: forward, left and Planning and Control (Advanced Textbooks in Control and Signal
right. The eye closing is used for wheelchair stop. Real-time Processing), Springer; 2nd Printing edition, 2011.
control and navigation results have presented on Cirrus Power [6] R. Gonzalez, and R. Woods, Digital Image Processing Using MAT-
Wheelchair. LAB, 2nd edition, Gatesmark Publishing, 2009.
[7] The MathWorks, inc., Image Processing Toolbox Users Guide, 2007.

2015 IEEE 7th International Conference on CIS & RAM 83