Frank Hoffmann, Visiting Researcher
University of Kiel
(Professor Lotfi A. Zadeh)
Funding Sources: BISC program, ARO-MURI DAAH04-96-1-0341 Integrated Approach to Intelligent Systems ,
Our project is concerned with intelligent systems used for the control of mobile robots. Within the scope of soft computing, our research focuses on an automatic design method for the knowledge base of a fuzzy system using genetic algorithms [1]. The main objective is to further develop these learning methods in regard to adapt fuzzy controllers for an autonomous robot. The usefullness of the controllers trained in a computer simulation is afterwards tested in real world experiments with a mobile robot (Fig.1)[2].
The first part of our research is concerned with a flexible, compact genetic representation of the knowledge expressed in the form of fuzzy rules. Our work includes the development of a new coding scheme as well as new genetic operators that guarantee a complete set of fuzzy rules free of contradictions. We are investigating the integration of reinforcement learning techniques in our approach which allow a more accurate evaluation of individual fuzzy rules and therefore speeds up the learning process.
The second part of our work focuses on dynamical and functional decomposition of the robot's behaviour into several primitive behaviours. The activation of these basic behaviours depends on the context defined by the given goals and the environmental situation of the robot. Our research is aimed at developing a methodology which not only adapts the simple behaviours, but which also learns how to coordinate them in a suitable way.