Pattern Formation by Collective Behavior of Competing Cellular Automata-Based Agents

We propose a novel game-theoretic multi-agent system approach to create a desired 2D pattern. We interpret a pattern formation problem as a variant of the iterated Spatial Prisoner’s Dilemma game, where evolutionary competing CA-based agents are used as learning machines. We design a payoff function reflecting a local goal of CA-based agent-players, and we show that the system of competing players is able to reach a Nash equilibrium, providing at the same time the maximization of a global criterion unknown for the agents that is related to the considered pattern formation problem. We provide experimental results showing a high performance of the pattern formation process.