Evolving Robot Controllers Using Carbon Nanotubes

Abstract

Evolution-in-materio uses computer-controlled evolution to configure materials to solve computational problems. In this paper, the material is a mixture of single-walled carbon nanotubes in an insulating polymer. We show for the first time that using purpose-built hardware it is possible to evolve voltages and signals applied to such materials to control robots, both simulated (Kephera) and real (Pi-Swarm). Evolved controllers were able to fully explore an environment, avoid obstacles, and cope with introduced faults, errors and environment changes. We also evolved a robot controller that could solve mazes of various difficulties.