Evolutionary predator-prey robot systems

Abstract

We present a feasibility study on evolving controllers for a group of wheeled robot predators that need to capture a prey robot. Our solution method works by evolving controllers in simulation for 100 generations, followed by 10 generations on real robots. The best controllers are further evaluated by their sensitivity for the initial positions. The results demonstrate the practical feasibility of this approach and give an indication of the time required to develop good solutions for the predator-prey problem.