Evolving Receptive-Field Controllers for Mobile Robots

Abstract

The use of evolutionary methods to generate controllers for real-world autonomous agents has attracted recent attention. Most of the pertinent research has employed genetic algorithms or variations thereof. Recent research has applied an alternative evolutionary method, evolution strategies, to the generation of simple Braitenberg vehicles. This application accelerates the development of such controllers by more than an order of magnitude (a few hours compared to more than two days). Motivated by this useful speedup, this paper investigates the evolution of more complex architectures, receptive-field controllers, that can employ nonlinear interactions and, therefore, can yield more complex behavior. It is interesting to note that the evolution strategy yields the same efficacy in terms of function evaluations, even though the second class of controllers requires up to 10 times more parameters than the simple Braitenberg architecture. In addition to the speedup, there is an important theoretical reason for preferring an evolution strategy over a genetic algorithm for this problem, namely the presence of epistasis.