SNAKE-LIKE BEHAVIORS USING MACROEVOLUTIONARY ALGORITHMS AND MODULATION BASED ARCHITECTURES

Abstract

In this paper we describe a methodology for obtaining modular artificial neural network based control architectures for snake-like robots automatically. This approach is based on the use of behavior modulation structures that are incrementally evolved using macroevolutionary algorithms. The method is suited for problems that can be solved through a progressive increase of the complexity of the controllers and for which the fitness landscapes are mostly flat with sparse peaks. This is usually the case when robot controllers are evolved using life simulation for evaluating their fitness. 1.