AN AGENT TEAM BASED REINFORCEMENT LEARNING MODEL AND ITS APPLICATION

摘要

Multi agent learning has attracted increasing attention in recent years. In this paper, a novel model for reinforcement learning based on agent team is proposed. Its basis is Q learning, a single agent reinforcement learning algorithm. The most significant characteristic of the model is the introduction of the active agent, the major role in team learning. By switching the active agent, team learning is achieved. A model in robotic soccer domain is implemented by extending the Q learing algorithm, and some positive results are obtained in experiments. Success in robotic soccer domain shows the effectiveness of the model.