CIMSA 2008 – IEEE International Conference on

Computational Intelligence for Measurement Systems And Applications

Istanbul – Turkey , 14-16 July 2008

An Occupancy Grid Based SLAM Method

Ozan Özışık, Sırma Yavuz
Yıldız Technical University
Computer Engineering Department

Abstract – Simultaneous localization and mapping (SLAM) is cycling environments. But there are some studies [4, 5, and 6]
an active area of research. SLAM algorithms should allow the robot that use IML while keeping track of the uncertainties, and
to start its movement from a random position in an unknown they can correct the map backwards in time whenever an
environment and to build the map of the area while knowing its own inconsistency is detected.
position relative to the map. Thus, at the end of the mapping task
robot should be able to return where it has started. Especially in
real time applications, using limited sensor data, there are still II. PROPOSED APPROACH
many problems to be conquered. In this study a probabilistic
occupancy grid approach is proposed to build the map of an In this study, a method that incrementally builds an
unknown environment. The method tested both in a simulation Occupancy grid map is proposed. The map is initialized with
environment and on a real robot. Although there are some an occupancy value in all the grids. The initial occupancy
improvements to be made, the initial results are promising. values are selected as 0.5 and are updated in every step that
sensor data is taken. Then the control signals are created
Keywords – SLAM, occupancy grid mapping, probabilistic robotics depending on this up to date map. In the map, occupancy is
represented by a rational number greater than 0 and in every
I. INTRODUCTION step that an obstacle is observed in a point’s neighborhood,
occupancy value of the point is increased. Occupancy update
Robotic Mapping has been a highly active research area is done according to equation (1). In this equation, P is the
for the last two decades [1]. Mapping is the process of the
point whose occupancy value will be updated, occ p is the
robot to model the environment it exists using the control
signals and the sensor data it gets. There is an important old occupancy value of P , occ p ' is the new occupancy
problem in application; sensor data is usually noisy and also
the robot may not be moving precisely as the control signals value of P , α is the degree of increment, X is the point
tell it to do. In order to solve this problem, probabilistic that an obstacle is observed, and R is a neighborhood
methods have been developed. Smith, Self and Cheeseman function that is inversely proportional to the distance between
developed a powerful statistical framework, for X and P .
simultaneously solving the mapping problem and the problem
of localizing the robot relative to its growing map, in 1990’s occ p ' = occ p * (1 + α * R (X, P)) (1)
[2,3]. Subsequent to this progress, robotic mapping has
commonly been referred to as simultaneous localization and
mapping (SLAM). As it is seen in the equation, the method doesn’t only
There are many SLAM algorithms, and Kalman Filter update the occupancy value of the point that an obstacle is
Method is probably the most commonly used. Kalman Filters observed, it also updates the occupancy values of all the
are Bayesian filters that represent posterior pose estimation points in a neighborhood. While the robot is moving, the
p(st,m | zt, ut) with Gaussian distribution. There are some points whose occupancy values are above a threshold are
drawbacks associated with Kalman Filter; it needs artificial considered as obstacles and decisions are taken depending on
landmarks, it can not solve the data association problem, these points.
sensor data has to be preprocessed, and noise in sensor data
must have a Gaussian distribution. Another popular algorithm A point’s occupancy value is updated in the following
is Expectation Maximization which is developed using the circumstances:
maximum likelihood model. As it doesn’t use an incremental 1) When an obstacle is observed in point P , P ’s
model while processing the data, the same data is processed occupancy value is increased.
again and again; the method doesn’t only depend on the last
2) When an obstacle is observed in point P , occupancy
step but it depends on all the previous steps. Expectation
maximization algorithm can solve the data association values of the points in the neighborhood of P are updated in
problem but it can not work in real time. Incremental a degree that is inversely proportional to their distances.
Maximum Likelihood (IML), is a method that creates the 3) When an obstacle is observed, occupancy values of the
map step by step while it gets the sensor data. As it doesn’t points between the obstacle and the robot are decreased.
consider the uncertainties, it is unsuccessful in mapping

978-1-4244-2306-4/08/$25.00 ©2008 IEEE

 4) When sensors don’t observe any obstacle, occupancy
values of the points that are on the direction of the sensors are
decreased.
5) Occupancy values of the points that are on the path that
robot is moving on are decreased.

III. STRUCTURE OF THE MOBILE ROBOT

The three wheeled mobile robot used in this study is

constructed in Computer Engineering Department of Yıldız
Technical University1.
While the SLAM problem may be considered as a solved
problem with sophisticated sensors, it is very challenging if
the robot has limited sensing capabilities. Therefore one of
the important properties of the robot used in this study is
being equipped with only 6 infrared sensors for proximity Figure 2 - The graphical user interface of the simulation environment
detection. Each of these sensors takes a continuous distance
reading and returns a corresponding analog voltage with a There is one important difference in simulation
range of 10cm to 80cm. environment which is that the sensor readings are ideally
An optical encoder is used to calculate the distance the correct or can only include a pre defined gauss or exponential
robot has traveled, based on the diameter of the wheel. The noise. While using the simulation environment where the IR
information obtained from IR sensors and optical encoder is sensor readings are assumed to be correct the proposed
used to localize the robot relative to its growing map. algorithm is able to produce correct map of the environment.
The robot is shown in Figure 1.

Figure 3 – Map for an L shaped corner obtained by using the

simulation environment

In second part of the experiments the algorithm is tested

Figure 1 - The three wheeled mobile robot used in the study on real robot, thus having noisy IR sensor readings. Some of
the results obtained for a rectangular room and an L shaped
IV. RESULTS corner are shown in Figure 4 and Figure 5.

Initially the proposed algorithm is tested on a simulation

environment to adjust the parameters. The simulation
environment is completely imitates the real robot including
the sensor readings and localization of the robot. The
graphical user interface of the simulation environment used in
our studies is shown in Figure 2. It is possible to place the
robot at any point to start with and create obstacles on desired
positions. The algorithm will control the simulation and it is
possible to follow the trajectory of the robot during the run.
The final map determined by the algorithm is exported to
an image file as seen in Figure 3.

1
This research has been supported by Yıldız Technical University Scientific Figure 4 - Map for an L shaped corner obtained by using the real
Research Projects Coordination Department. Project Number: 27-04-01-01. robot
Figure 5 - Map for a rectangular area obtained by using the real robot

As it can be seen from the figures, light gray grids that are
inconsistent with the trajectory of the robot are the result of
noisy sensor readings. These noisy data may lead to wrong
identification of obstacles (walls). As a solution currently it is
planned to calibrate the sensors via software to remove the
sensitivity differences between them. Upon experiments also
the threshold values will be determined more accurately.

REFERENCES

[1] S. Thrun, (2002), Robotic mapping: A survey, Exploring Artificial

Intelligence in the New
Millenium, Morgan Kaufmann.
[2] R. Smith, M. Self, and P. Cheeseman. Estimating uncertain spatial
relationships in robotics. In I.J. Cox and G.T. Wilfong, editors,
Autonomous Robot Vehnicles, pages 167–193. Springer-Verlag, 1990.
[3] R. C. Smith and P. Cheeseman. On the representation and
estimation of spatial uncertainty. Technical Report TR 4760 & 7239,
SRI, 1985.
[4] J.-S. Gutmann and K. Konolige. Incremental mapping of large
cyclic environments. In Proceedings of the IEEE International
Symposium on Computational Intelligence in Robotics and Automation
(CIRA), 2000.
[5] S. Thrun. A probabilistic online mapping algorithm for teams of
mobile robots. International Journal of Robotics Research, 20(5):335–
363, 2001.
[6] S. Thrun, W. Burgard, and D. Fox. A real-time algorithm for mobile
robot mapping with applications to multi-robot and 3D mapping. In
Proceedings of the IEEE International Conference on Robotics and
Automation (ICRA), San Francisco,
CA, 2000. IEEE.
[7] S. Thrun, W. Burgard, D. Fox, Probabilistic Robotics, The MIT
Press, 2005.