Intelligent control in dynamic system

This paper presents a method for obtaining optimal policies which allow for action planning and optimal control in a non-communicating multiagent system. In this system homogeneous agents have the same structure and domain but act and are situated differently in the world. The lack of information about each other's internal state and observation inputs may lead to non-desirable prediction in planning and control, since they may not be able to predict the world change. To cope with missing knowledge, agents simulate each other's behavior as part of the environment dynamic and build their own policy based on local data from the simulation episodes. A reinforcement learning method is applied to derive the policy of future actions, where agents compute and update the result repeatedly towards the goal. The Monte Carlo approach is used in solving the reinforcement problem from simulated experiences. Since each agent learns to adapt its policy to environment changes, the global picture is supposed to appear as multiagent coordination. The multiple vehicles domain, where there is no communication among the vehicles and sensing of vehicles is limited, is considered under a simulation model.