Soft computing techniques for the design of mobile robot behaviors

This paper describes a soft computing approach to robot behavior design. Evolutionary algorithms adapt a wall following behavior implemented by means of fuzzy control rules, A messy coding scheme for fuzzy rules reduces the size and complexity of the rule base, with the result, that the fuzzy control design remains tractable for the genetic algorithm. An evolution strategy tunes the scaling factors for the controller's input and output and optimizes the adjustment of the sensors on the robot. A neural network classifies the basic geometric features of the local environment based on previous perceptions of the sonar sensors. (C) 2000 Elsevier Science Inc. All rights reserved.