Journal of Electrical Engineering

www.jee.ro

OBSTACLE AVOIDANCE AND CONTROL OF SWEDISH WHEELED

MOBILE ROBOT
Mhuammad ADIL, AhsanALI, Muhammad A. CHOUDHRY

University Of Engineering and Technology Taxila, Pakistan

Email: [email protected]

Abstract: The main focus of this work is on obstacle

avoidance and trajectory tracking of Swedish
wheeled omnidirectional mobile robot. The proposed
obstacle avoidance algorithm takes information from
its onboard sensors and modulates trajectory such
that the robot moves towards the goal, having
maximum distance from the obstacles and minimum
deviation from the trajectory. The proposed
trajectory tracking controller is formulated based on
linear model predictive control (MPC). The
motivation of the use of linear MPC is its less
computational cost relative to the nonlinear MPC.
Realistic simulations are performed to test the Figure 1: Swedish Wheel
Figure 2: Swedish Wheeled
validity and performance of the proposed obstacle omnidirectional WMR
avoidance algorithm and the trajectory tracking
controller.
Key words: Trajectory Tracking Linear Model
Predictive Control, Obstacle Avoidance, Swedish
wheeled omnidirectional WMR.

1. Introduction
A standard wheel with small rollers mounted on
its perimeter is called Swedish wheel. It has an extra
degree of freedom to the fixed standard wheel. It was
invented by BngtIlon in 1973 [1]. The angle
between the rollers axis and the wheel hub axis
direction is typically equal to or . The wheel
Figure 1: Navigation Architecture of WMR
with is not used, because it does not provide
extra degree of freedom to the wheel. Fig. 1 is the truly omnidirectional, i.e. it can move in any
picture of the Swedish wheel . Fig. 2.is the direction without reorienting itself. The robot has 16
picture of 3WD omnidirectional mobile robot (a sonars for obstacles detection and a dioptric vision
creation of IdMind - Engenharia de Sistemas, Lda) system for localization. The architecture of this robot
with three Swedish wheels, each wheel is apart is based on a central processing unit, a notebook PC.
from the other. This configuration makes the robot

1
 Journal of Electrical Engineering
www.jee.ro

This unit gathers information from all other sub- With such that the control input:
systems and sensors. The robot uses three Maxon DC
15V 90W motors for locomotion, with a 21:1 gear derives error to zero:
relation and digital encoders. Each motor is
controlled by a Faulhaber MCDC2805 controller.
The MCDC2805 receives speed, acceleration and The desired reference velocity or tracking error
position commands through its RS232 port, and can always happen to be large enough for the
interfaces directly with the motor and encoders. To actuator to reach their limits. And the Proportional
control the motors by USB, there is an electronics control law cannot easily handle actuator constraints.
board on the robot that translates USB signals from We have proposed linear Model Predictive Control
the PC to a standard RS232 serial signal to the (LMPC) for trajectory tracking, which can efficiently
controller. handle actuator constraints. The general design
objective of model predictive control is to compute a
Fig. 3 is the overall navigation architecture of trajectory of a future manipulated variable to
WMR that is followed in this research work. In the optimize the future behavior of the plant output .
path planning step robot assumes global knowledge The optimization is performed within a limited time
of the environment to decide what to do over the long window by giving plant information at the start of the
term to achieve its goal [2]. Road Map [3, 4], cell time window. Low level control is designed in a
decomposition [5], potential field [6] are some of the decentralized fashion. Dynamic coupling among the
strategies used for path planning. Obstacle avoidance actuators are neglected, and each motor is controller
focuses on modulating the robot’s trajectory as separately, with a velocity PID loop to follow the
informed by its onboard sensors during its motion, so speed command from inverse kinematics (Faulhaber
that collision with obstacles is avoided. Bug MCDC2805 controller).
algorithm is the simplest obstacle avoidance
technique for WMRs [7]. The bug algorithm Rest of paper is divided into four sections. In
guarantees completeness, but trajectory generated by section 2, mathematical model of the omnidirectional
bug algorithm is very inefficient. According to the WMR is developed and validated. In section 3,
bug algorithm the robot has to fully encircle the proposed new obstacle avoidance algorithm, is
obstacle first, then it departs from the point with the presented. In section 4, the developed trajectory
minimum distance towards the goal. In this paper we tracking controller is written. And finally, the
proposed an obstacle avoidance algorithm that simulation results and conclusion is written in section
generates shortest trajectory with minimum control 5 and 6 respectively.
effort. The proposed algorithm also contains tuning
parameters that can be used for performance
optimization.
The trajectory tracking control problem consists
in the stabilization of error e (with respect to the
position of moving reference robot) to zero. In the
open loop control, the robot is not able to
automatically correct or adapt the trajectory if a
dynamic change of the environment occurs. So
feedback controller is more appropriate approach in
motion control of a mobile robot. Proportional
control law presented in [8, 9] can be used to design
trajectory tracking controller for omnidirectional
WMR. The task of proportional control law is to find
out matrix K, if it exists:

2
 Journal of Electrical Engineering
www.jee.ro

2. Mathematical Modeling
Kinematic analysis of omnidirectional WMR has
been addressed in several papers[10-12]. Dynamic
model of omnidirectional WMR has also been Where are the platform center velocities in
developed in [13], but dynamic model of the the direction of x-axis and y-axis respectively, is
omnidirectional WMR is not very common because the angular velocity and M is a matrix composed of
of the difficulty in modeling the several internal (unit vector parallel with roller axis) and
frictions inside the wheel. Kinematic model (tangent vector direction of the roller).
discussed below is very similar to the one proposed
by Giovanni Indeveri [14]

2.1. Kinematic Model 1:

Mapping of robot’s motion in global RF in terms
of local RF can be achieved by coordinate For controllability should be equal to 3,
transformation [2](fig. 4). and should not be equal to 0. In case of 3WD
robot we have = 0, and each wheels are apart
from the other. So we can write:

Where R is the radius of wheel and is the distance

between the center of the robot and wheel.

2.3. Empirical Model Validation

To verify the model, the kinematic model 2 is
used to demonstrate whether the robot moves as
expected. An overhead camera is used to observe the
motion of 3WD robot. Fig. 5,6 and 7 shows the
output of model (blue line) and the path followed by
the 3WD omnidirectional WMR (red line) for the
ramp, sinusoidal, and circular trajectories
respectively. From the figures it is obvious that the
error between model and the robot’s output is
negligible. To demonstrate the model validation
Figure 2: Coordinates transformation results more quantitatively we have repeated the
experiment for different velocities and calculated
2.2. Kinematic Model 2: mean square error (between 3WD robot’s output and
the developed mathematical model output). Table II
Kinematic model of the WMR is a relationship is the summary of model validation experiments.
between the robot speed , the wheels speed ,
and the configuration coordinates (geometric
parameters of a robot). The kinematic model
developed is for general setting of N number of
Swedish wheels with arbitrary (but fixed) roller
wheel angle can be written as [15].

3
 Journal of Electrical Engineering
www.jee.ro

3. Obstacle Avoidance
A new method of obstacle avoidance is proposed
which consists of three steps. These three steps are
repeated at each sampling instant, until the robot
reaches its destination. In the first step local
environment of the robot is scanned, in the second
step a local map is created which contains
information about obstacle and goal locations, and in
the third step desired new position of the robot for
the next sampling instant is calculated. The
Figure 3: Ramp Trajectory calculated new position is the most optimal in the
sense that it has maximum distance from the
obstacles, minimum distance from the goal point, and
change in control input is also minimum. Like the
most of other obstacle avoidance techniques, we have
considered the robot as a point capable of holonomic
motion. Ultrasonic sensors are used for scanning
local environment of the robot. The sensors have
range between 12 cm to 5 m and accuracy of about 2
cm [16]. Ultrasonic sensors inherently suffer from
several drawbacks, namely bandwidth and cross-
sensitivity. But we improved the bandwidth by using
Figure 4: Sinusoidal Trajectory a ring of ultrasonic sensors. To incorporate
inaccuracies of sensors data obstacles are represented
in a probabilistic fashion. A grid type model of
robot’s local environment is developed, in which
each cell in the grid has certainty value (CV)[16]. CV
is the measure of confidence that an obstacle exists
within the cell area. Cells having CV less then
threshold are ignored and assumed to be free of
obstacles. Certainty grid is developed using same
method that is proposed in well-known research
work, titled ’’virtual field histogram”[17]. Once the
certainty grid has been constructed, the obstacle
Figure 5: Circular Trajectory avoidance algorithm needs to move the robot in such
a way that obstacles are avoided and the robot
proceed towards the target. The robot’s linear
velocity magnitude can be set at the beginning of run.
And direction of the velocity is chosen such that it
minimizes the objective function given by:

Where is certainty value of obstacle , is

distance from the goal, is distance from the
obstacle , and is the change in control input.
Change in control input is calculated using kinematic
model 1 of the WMR. First term in the cost function
penalizes the change in control input Δ while the
second term penalizes the distance from the goal.

4
 Journal of Electrical Engineering
www.jee.ro

Third term is used to maximize the distance of a

robot from the obstacles. P and Q are weighting
matrices that are used for tuning purpose. In the fig.
8, dotted circle – – shows
possible next locations of the robot for next sampling
instant. Where
.

Objective function is evaluated at 360 points on the

circle, and the point at which the objective function
gives minimum value is chosen to be the desired
location of robot for the next sampling instant. To
test the performance of proposed obstacle avoidance
algorithm simulations were made using MATLAB
with the assumption that we have perfect position
information i.e. sensors or actuators noise is
neglected. Some of the simulation results are given
below (Fig. 9.), which shows feasibility of the
proposed technique. Tuning parameters Q and P can
be used to further optimize the proposed algorithm to
obtain the desired behavior of robot.

Figure 7: Simu lation Results

4. Trajectory Tracking
In this section we have proposed a linear model
predictive control approach for trajectory tracking of
Figure 6: Obstacle Avoidance omnidirectional WMR. Although nonlinear model
predictive control approach for trajectory tracking of
WMR have already been proposed in literature [18],
but nonlinear MPC have much more computational
load than the linear MPC approach, so linear MPC is
more suitable for fast moving robots. The proposed
technique is similar to that proposed in [19] for
trajectory tracking of nonholonomic WMR but this
technique has not been used for trajectory tracking of
holonomic (omnidirectional) WMR (according to the

5
 Journal of Electrical Engineering
www.jee.ro

author’s best knowledge). First step of the proposed

Linear MPC trajectory tracking controller is to find
out a linear time varying description of the system.
Then, taking control input as a decision variable, the
optimization problem to be solved at each sampling
instant is transformed into QP problem. Now QP
problem could be solved by numerically fast and
robust algorithms which lead to global optimal
solutions.
Kinematic Model 1, which is simply the
transformation of robot’s motion form local RF into
the global RF is successively linearized at the
trajectory points and .

It can be supposed that given trajectory is generated

by a virtual robot, which has same model as the
omnidirectional WMR. We can write the linearized
dynamic model of the robot as below:

Error dynamics can be written as :

Where and . Given the

sampling time, we can obtain discrete time model of
the robot by using forward difference approximation.

We can write the objective function as:

Where Q, and R are the weighting matrices. Now we

can write as below [21]:

Let as given below.

6
 Journal of Electrical Engineering
www.jee.ro

Now we can write the objective function in standard

QP from.

With

Following optimization problem is solved at each

sampling time
Figure 8: Circular Trajectory

s.a

Amplitude constraints can be written in terms of

decision variable as:

Where and are upper and lower limits of Figure 9: Error Dynamics
the deviation from the reference robot’s speed,
respectively.

5. Simulation Results
We have performed realistic simulation of Linear
Model Predictive Trajectory Tracking controller
using MATLAB 2010b. The QP problem is solved
online using MATLAB QP solver (quadprog). The
average computation time is 3 ms. Fig 10. shows the
simulation results of circular trajectory tracking.
Where blue line stands for the reference trajectory
and red line represents the trajectory of WMR by
using linear MPC algorithm. It is clear that the robot
Figure 10: Control Inputs
moves toward the trajectory and continue tracking it.
We have used Np=20, Nc=5, P=I,Q=I, these 6. Conclusion
parameter can be tuned according to the desired
performance criteria. Fig. 11 shows that error In this paper an obstacle avoidance algorithm and
(vertical, horizontal, and angle errors) converges to trajectory tracking controller law for omnidirectional
zero. Figure 12 is the control input to each wheel, WMR is presented. A nonlinear model of the
control inputs are calculated using inverse kinematics omnidirectional WMR is developed and validated
(kinematic model 2). The generated control input is experimentally, then linear time varying MPC is
very much smooth and within the actuator’s limits. formulated for the trajectory tracking of the robot. In

7
 Journal of Electrical Engineering
www.jee.ro

the proposed controller optimization problem to be Wheels." Journal of Robotic Systems 21.4 (2004):
solved is transformed in QP problem, which can be 193-208.
solved using numerically fast and robust algorithm. 13. Liu, Yong, et al. "Omn i-directional mobile robot
Both trajectory tracking and obstacle avoidance controller based on trajectory linearization." Robotics
algorithm are verified by performing simulations. and Autonomous Systems 56.5 (2008): 461-479.
14. Indiveri, Giovanni. "Swed ish wheeled
The simulation results showed the flexibility and omnid irectional mobile robots: kinematics analysis
effectiveness of the proposed techniques. and control." IEEE transactions on robotics 25.1
(2009): 164-171.
15. Everett, H. R. Sensors for mobile robots: theory and
application. AK Peters, Ltd., 1995.
7. References 16. Borenstein, Johann, and YoramKoren. "The vector
1. Diegel, Olaf, et al. "Improved mecanum wheel design field histogram-fast obstacle avoidance for mobile
for o mni-directional robots." Proc. 2002 Australasian robots." Robotics and Automation, IEEE Transactions
Conference on Robotics and Automation, Auckland . on7.3 (1991): 278-288.
2002. 17. Kuhne, F., Walter Fetter Lages, and J. M. Go mes da
2. Siegwart, Roland, Illah Reza Nourbakhsh, and Silva Jr. "Point stabilization of mobile robots with
DavideScaramu zza. Introduction to autonomous nonlinear model predict ive control."Mechatronics and
mobile robots. MIT press, 2011. Automation, 2005 IEEE International Conference.
3. Kalmár-Nagy, Tamás, RaffaelloD’Andrea, and Vo l. 3. IEEE, 2005.
PritamGanguly. "Near-optimal dynamic trajectory 18. Künhe, F., J. Go mes, and W. Fetter. "Mobile robot
generation and control of an omnid irectional trajectory tracking using model predict ive control." II
vehicle."Robotics and Autonomous Systems 46.1 IEEE latin-american robotics symposium. 2005.
(2004): 47-64. 19. Raffo, Gu ilherme V., et al. "A predictive controller
4. Wu, Jianhua. Dynamic path planning of an omni- for autonomous vehicle path tracking." Intelligent
directional robot in a dynamic environment. Diss. Transportation Systems, IEEE Transactions on 10.1
Ohio University, 2005. (2009): 92-102.
5. Lingelbach, Frank. "Path planning using probabilistic
cell decomposition."Robotics and Automation, 2004.
Proceedings. ICRA'04. 2004 IEEE International
Conference on. Vo l. 1. IEEE, 2004.
6. Barraquand, Jerome, Bruno Langlois, and J-C.
Lato mbe. "Nu merical potential field techniques for
robot path planning." Systems, Man and Cybernetics,
IEEE Transactions on 22.2 (1992): 224-241.
7. Ng, James, and Thomas Bräunl. "Perfo rmance
comparison of bug navigation algorithms." Journal of
Intelligent and Robotic Systems 50.1 (2007): 73-84.
8. Astolfi, A lessandro. "Exponential stabilizat ion of a
wheeled mobile robot via discontinuous
control." Journal of dynamic systems, measurement,
and control121.1 (1999): 121-126.
9. Lee, Sung-On, et al. "A stable target-tracking control
for unicycle mob ile robots." Intelligent Robots and
Systems, 2000.(IROS 2000). Proceedings. 2000
IEEE/RSJ International Conference on. Vol. 3. IEEE,
2000.
10. Samani, A. H., et al. "Design and development of a
comprehensive omni direct ional soccer player
robot." International Journal of Advanced Robotic
Systems 1.3 (2004): 191-200.
11. Muir, Patrick F., and Charles P. Neu man. "Kinematic
modeling for feedback control of an o mnid irectional
wheeled mobile robot." Autonomous robot vehicles.
Springer New Yo rk, 1990. 25-31.
12. Song, Jae‐Bok, and Kyung‐SeokByun. "Design and
Control of a Four‐Wheeled Omn idirectional Mobile
Robot with Steerable Omnid irectional