Reinforcement learning in zero-sum Markov games for robot soccer systems

The objective of this paper is to develop a strategy system in a robot soccer system with cooperative ability which is improved by self-learning. A reinforcement learning method according to the zero-sum game theory is developed in this paper. It enforces the learning systems to choose appropriate strategy on the opponent´s actions. In order to achieve the purpose of cooperation, two sub systems have been used, one is a role assignment system and the other one is a reinforcement learning system.