LABORATOIRE DE MICRO-INFORMATIQUE - EPFL
| Neuro
- Fuzzy Controllers http://diwww.epfl.ch/lami/robots/fuzzy.html Jelena Godjevac
|
LAMI - Mantra Day
Microprocessor
Education
December 5th, 1997
tems. Their main goal is to implement human know-how or heu-
ristic rules in the form of a computer program. Fuzzy logic
provides a mathematical formalism for this goal. An original
method for the design and analysis of a fuzzy controller is pro-
posed. The main idea is to apply an on-line adaptive algorithm to
automatically adjust the parameters of the fuzzy controller,
depending on its inputs and desired outputs. It is shown that a
fuzzy controller can learn to approximate non-linear functions
arbitrarily well. A new method for the extraction of linguistic rules
from the adapted parameters is introduced. The knowledge
acquired during learning can be represented in a clear and intui-
tive syntax. This gives the designer a comprehensive under-
standing of the controller features.
The application of fuzzy controllers to the navigation of a mobile
robot is investigated. The proposed design method for fuzzy con-
troller is tested, showing that it provides the designer with a tech-
nique to handle complex tasks: the robot can successfully learn
the desired behaviors, even with noisy sensors and large rule
bases. The extraction of rules reveals to be extremely useful for
practical applications because it helps the engineer to learn more
on the complete system.
platform for the teaching of fuzzy control theory and neuro-
fuzzy networks for postgraduate students and programming
languages for graduate students.
Khepera makes teaching of fuzzy-control theory easier and
more compelling for students. They learn how to implement
a simple fuzzy control algorithm for the task of obstacle
avoidance. Students have to build a fuzzy controller step-by-
step using the software package LabVIEW. It allows them to
understand basic fuzzy logic concepts, the advantages and
drawbacks of fuzzy controllers as well as the difficulties of
their practical design.
Khepera is also a very good tool for teaching programming
languages. Microengineering students program the robot to
accomplish a specified task. This help them to become famil-
iar with the field of mobile robotics and to learn programming
languages while having fun.
Fuzzy
controller for Khepera (LabVIEW)Jelena Godjevac
received
the degree in Electrical Engineering from Belgrade University, in1989, and the doctoral degree from the Swiss Federal Institute of Technology of Lausanne in
1996. Her doctoral thesis was published as a book by Presses Polytechniques Universitaires
Romandes in 1997. She is a lecturer and researcher at LAMI. The main direction of her research
is the application of neuro-fuzzy algorithms in the field of mobile robot control. Currently, she is
preparing her second book on fuzzy logic and neuro-fuzzy controllers for publication in 1998.
E-mail: Jelena.Godjevac@epfl.ch