Measuring emergent behaviors in a mixed competitive-cooperative environment

Abstract

A fundamental challenge in evolutionary theory is explaining the evolution of cooperative (or altruistic) tendencies despite local competition among agents can limit cooperative benefits. In this paper, an agent-based model is developed that combines network evolution strategies with conflicting pressures to induce cooperation as an emergent behavior. Thoroughly, we define a model of two agents able to evolve cooperative actions in a mixed competitive-cooperative environment. Specifically, two simulated E-puck robots are put inside an arena filled with diverse kinds of food items (i.e., individual, social). The goal is to survive as long as possible by eating food to contrast energy consumption. Robot controllers, which determine the agent’s interaction with other agents, are evolved by using a genetic algorithm. Simulation results suggest that by side with expected behaviors, a new strategy emerges without any external pressure. Outcomes allow conclusions about the feasible cooperation choices individuals should make when participating in complex mixed cooperative-competitive scenarios. In particular, we observe a natural emergence of opportunistic behaviors in agents when such strategies can lead to the team’s success.