Dr Margaret Jefferies
Computer Science Department

Last year the Artificial Intelligence Laboratory acquired a Real
World Interface B14r robot on which to test and develop our cognitive
mapping theories. The basic idea is that the robot will acquire its
own navigational map from its experience as it moves around its
environment. Errors from wheel slippage and sensing are a huge
problem for robots. As the robot moves around its environment these
errors quickly accumulate so that robot's location in its map no
longer matches its location in its physical environment. This is the
robot localisation problem, how to accurately specify where the robot
is in its environment. The current popular approaches to this problem
focus on reducing the error to an acceptable level. By contrast we
argue that the focus should be on finding a suitable structure for
the map within which the location of the robot can be adequately but
not necessarily accurately specified. This structure has been
outlined in our cognitive mapping theories in which we argue that the
basic element should be the space which encloses the robot - we call
it the ASR. It is the ASR defines whereabouts the robot is within its
larger environment. This seminar will present our preliminary results
in constructing such a representation for a robot which uses sonar
to sense its environment.
--
Dr. Margaret Jefferies Tel: +64-3-4798584
Artificial Intelligence Laboratory Fax: +64-3-4798529
Department of Computer Science email: megan@cs.otago.ac.nz
University of Otago
Dunedin
New Zealand