Evolution for behavior selection accelerated by activation/termination constraints

Abstract

This paper proposes an evolutionary approach to select appropriate behaviors for a mobile robot. We augment each behavior by adding activation/termination constraints to accelerate the evolutionary processes. The former constraints reduce the number of situations where each behavior is executable, and the latter contribute to extract meaningful behavior sequences, each of which can be regarded as one action regardless of its length. We apply the genetic algorithm to obtain the switching function to select the appropriate behavior according to the situation. As an example, a shooting task in a soccer game is given to show the validity of the proposed method. Based on the combination of the proposed architecture and GA, we can obtain the purposive behaviors. Simulation results are shown, and a discussion is given.