In Proceedings of the International Conference of Machine Learning and Cybernetics 2011 (ICMLC 2011)
Abstract:In multi-agent pursuit-evasion game, pursuers need to coordinate the behavior of each other to achieve a common goal: catching the evader. In this paper, a learning mechanism to capture evader in a dynamic environment of pursuit-evasion game is proposed. It deals with the uncertainty in environment using training and, according to whether the agents cooperated with each other, agent learning is divided into two ways: one is individual learning and the other is case-based reasoning. With these methods, agents are capable of memorizing and learning in order to catch evader more quickly.