Online learning of fuzzy behaviours using genetic algorithms and real-time interaction with the environment

This paper addresses the development of an online self-learning control system for mobile robots. In particular we describe a novel fuzzy logic and genetic algorithms (GA) method to implement a behaviour-based architecture that has significant design, analysis and performance advantages over previous approaches. We describe a new algorithm to learn fuzzy behaviours where reinforcement can be given as actions being performed. A modified version of the fuzzy classifier system (FCS) is used. The FCS is equipped with a rule-cache making it possible for the learnt expertise to be applied to future situations and to allow GA learning to start the search from the best point found. The system uses sensory information in-order to narrow the search space for the GA. The proposed techniques have resulted in rapid convergence suitable for learning individual behaviours online without need for simulation. The results of this work are compared with results reported elsewhere and reveal this approach has a superior performance to existing systems.