Coevolution of Competing Agent Species in a Game-Like Environment

摘要

Two species of agents coevolve in a 2D, physically simulated world. A simple fitness function rewards agents for shooting at agents of the other species. An evolutionary framework consisting of the grid-brain agent controller model and the SEEA steady-state evolutionary algorithm is used. We were able to observe a phenomenon of species specialization without the need for geographical separation. Species with equal initial conditions were shown to diverge to different specialization niches by way of the systems dynamics. This kind of research may lead to more interesting gaming environments, where the world keeps changing and evolving even in the absence of human interaction.