Evolving neural controllers for visual navigation

Biological evidence strongly suggests that insects utilize visual cues for their navigation tasks. This paper discusses the evolution of a simple controller for visual homing by means of evolutionary algorithms. The application is representative for a class of (real world) problems, for which the choice of the fitness function is non-trivial, since the data are not known in advance. For this class of problems, recombination has a much higher influence on the convergence than previously assumed. We show how convergence rates comparable to those of neural network learning algorithms can be achieved