Behavior Learning of a Group of Mobile Robots with a Distributable Genetic Algorithm.

Abstract

The purpose of this paper is to build an autonomous distributed robot system composed of mobile robots of the same type with the mobility of micro-mice and with a simple communication apparatus. The feature of the proposed robot system is its behavior learning capability by a distributable genetic algorithm (dGA) . dGA differs from GA in the treatment of fitnesses of individuals and can be implemented on a group of mobile robots. The robot behaviors are modeled by simple finite automata and their transition functions are encoded into genes in each individual robot. The transition functions determine the performance of the robots and are subject to genetic operations according to their results with a generation in dGA as a trial in the learning process. A simulation with 100 individuals (4-state 2-input automata) verifies the validity of the proposed approach. The behavior learning is found to improve the performance of the robot system as a whole.