Abstract. This paper explores the application of an artificial developmental system (ADS) to the field of evolutionary robotics by investigating the capability of a gene regulatory network (GRN) to specify a general purpose obstacle avoidance controller both in simulation and on a real robot. Experiments are carried out using the e-puck robot platform. It is further proposed to use cross-correlation between inputs and outputs in order to assess the quality of robot controllers more accurately than with observing its behaviour alone.
Martin Trefzer, Tüze Kuyucu, Julian F. Miller