Intelligent world and systems were a dream in the past,but by growing artificial intelligence field it in becoming a reality. the main factor in a intelligent system is learning and artificial intelligence makes it possible. mutiagent system learning becomes more accurate and faster by combining machine learning methods with them. multi agent learning include Cooperative and Competitive methods. In Competitive learning agents try to increase their utility however other’s utility may be decreased. In cooperative learning agents try to increase utility of all agents simultaneously. In recent years many works have been performed in cooperative learning, Most of these methods used reinforcement learning for learning. While these methods have a main challenge in how to combine the knowledge of agents. In this thesis we have addressed some of those challenges to improve Cooperative learning methods. to achieve this objective , some main points of cooperative learning have been detected, the first point is action selectionin reinforcement learning which we used a new heuristic function to select actions, the second and third points are combining knowledge and task division by two criteria “shortest experienced path” and “Shock”. By using these two criteria combine the knowledge has been improved. Overall experiments showed improve in quality and learning speed. Key Words: Cooperative learning, Multi-agent system,Reinforcement learning
