Cooperative learning and planning for multiple robots

Abstract

The paper deals with the the subject of learning and planning for real mobile robots, using Sutton's (1991) Dyna algorithm. The Dyna algorithm integrates reinforcement learning, planning and reactive execution. We present an extension of the Dyna algorithm which includes symmetric and cooperative learning with multiple robots. We applied the extended version of the algorithm to a population of two real robots. Practical problems associated with the implementation of the algorithm on a real setup are solved. Results obtained from simulations and real experiments are presented and discussed.