Spatial Cognition and Computation 1: 205–226, 1999.
© 2000 Kluwer Academic Publishers. Printed in the Netherlands.
Using fuzzy sets to represent uncertain spatial
knowledge in autonomous robots
and ALESSANDRO SAFFIOTTI
IRIDIA, Université Libre de Bruxelles, 50 av. F. Roosevelt, CP 194/6, B-1050 Brussels,
Belgium (e-mail: email@example.com);
Applied Autonomous Sensor Systems, Dept. of
Technology, University of Örebro, S-70182 Örebro, Sweden (e-mail: asafﬁo@aass.oru.se)
Abstract. Autonomous mobile robots need the capability to reason from and about spatial
knowledge. Due to limitations in the prior information and in the perceptual apparatus, this
knowledge is inevitably affected by uncertainty. In this paper, we discuss some techniques
employed in the ﬁeld of autonomous robotics to represent and use uncertain spatial knowledge.
We focus on techniques which use fuzzy sets to account for the different facets of uncertainty
involved in spatial knowledge. These facets include the false measurements induced by bad
observation conditions; the inherent noise in odometric position estimation; and the vagueness
introduced by the use of linguistic descriptions. To make the discussion more concrete, we
illustrate some of these techniques showing samples from our work on mobile robots.
Key words: environment modeling, fuzzy logic, linguistic descriptions, robot navigation, self
localization, spatial maps, uncertainty management
Current research on autonomous mobile robots aims at building physical
systems that can move purposefully and without human intervention in
natural environments – that is, real-world environments that have not been
speciﬁcally engineered for the robot. Despite the recent advances in this
ﬁeld, autonomous robot navigation still requires a better understanding of
the processes involved in the perception and representation of space. The use
of general purpose sensors that are ﬁxed on the robot, the lack of structure in
these environments, and the need to ﬁnd solutions that do not depend on one
speciﬁc environment or task, impose strong limitations on the information
that will be available to the robot.
Most of these limitations originate in the nature of real-world, natural
environments. First, prior knowledge about the environment is, in general,
A preliminary version of this paper was presented at the Mind III Conference, Dublin,
Ireland, August 1998.