An empirical evaluation of evolutionary controller design methods for collective gathering task

Abstract

This research aims to evaluate the performance of evolutionary controller design methods for developing a collective behaviour for a team of robots. The methods tested in this research are NEAT which is capable of finding minimal solution quickly, and SANE which maintains high genetic diversity through neuron level evolution. The task chosen for these methods was a collective gathering task which required a team of robots to cooperate in finding and retrieving item of interest. Our results showed that NEAT consistently produced better controllers compared to SANE.