Increasing Adaptivity through Evolution Strategies

Abstract

The automated generation of controllers for real-world autonomous agents by means of evolutionary methods has recently attracted a lot of attention. Most of the pertinent research has employed genetic algorithms or variations thereof. We have applied a different evolutionary method to the generation of a control architecture for Braitenberg vehicles, namely "evolutionary strategies". The application of the ES accelerates the development of such controllers by more than one order of magnitude (a few hours compared to more than two days). This result is very important, since the development process is to be done in real systems. In addition to the dramatic speedup, there is an important theoretical reason for preferring evolutionary strategy over genetic algorithms, namely epistatic interaction. 1 Introduction Autonomous agents are self-sufficient, embodied systems (robots) [5] that do some useful work. Autonomous means that the agent operates without any human control. Self-sufficient...