Abstract:A cooperative team of agents may perform many tasks better than isolated agents. The question is how cooperation among self-interested agents may be achieved. It is important that, while we encourage cooperation among agents to form a team, we maintain autonomy of individual agents as much as possible, so as to maintain flexibility and generality. This paper presents an approach toward this goal, based on bidding utilizing reinforcement values acquired through reinforcement learning. The result is a simple and straightforward method that is generic and works in a variety of task domains. We further apply evolutionary computation to enhance cooperation among agents of a team, through selecting and reproducing those teams that are able to cooperate. We tested and analyzed this approach, MARLBS, in a variety of task domains, and demonstrated that a team of self-interested agents indeed performed better than the best single agent as well as the average of the single agents. In particular, Backgammon players trained using this approach outperformed PubEval (a publicly available benchmark player). These results validated our approach.