A Fuzzy-Neural Realization of Behavior-Based Control Systems for a Mobile Robot

Abstract

A fuzzy-neural realization of a behavior-based control system is described for a mobile robot by applying the soft-computing techniques, in which a simple fuzzy reasoning is assigned to one elemental behavior consisting of a single input-output relation, and then two consequent results from two behavioral groups are competed or cooperated. For the competition or cooperation between behavioral groups or elemental behaviors, a suppression unit is constructed as a neural network by using a sign function or saturation function. A Jacobian net is introduced to transform the results obtained from the competition or cooperation to those in the joint coordinate systems. Furthermore, we explain how to learn the present behavior-based control system by using a genetic algorithm. Finally, a simple terminal control problem is illustrated for a mobile robot with two independent driving wheels.