Simulating Morphological Evolution in Large Robot Populations

Abstract

Computational capacity and memory are limiting factors when simulating large numbers of robots with complex bodies: available physics engines struggle to handle more than a couple of dozens of complex robot bodies. This limits the possibilities of investigating the evolution of robot morphology to small populations with few generations. We present a method to simulate large evolving populations of robots with complex and varying morphologies. By simulating individual robots in parallel, we sacrifice the possibility of interaction between robots (other than to exchange genomes), but gain the opportunity to simulate substantial populations, not so much limited by the capabilities of the simulator itself as by the number of processors at our disposal.