p Zahadat - Msc student and Corresponding author
S.D Katebi - Faculty of Computer Science and Engineering, Department of Computer science and Engineering, School of Engineering, Shiraz University, Shiraz, Iran

In this paper a reinforcement Learning (RL) methodology is used to design multi agent systems. Our agents have to coordinate to reach a common goal. We used simple reactive agents with local perception in a decentralized way to overcome difficulties of RL. To increase speed of learning, we used an incremental learning algorithm where agents first face simple tasks to learn then face more and more hard tasks while keep learning.