Learning On Multi-Agent System

Reinforcement learning(RL) has been applied on the testbed of keepaway, a typical multi-agent system. To get the best performances of RL and figure out what influence the learning speed and the finally results, different values of the important variables had been selected in the emulator. The results intimated the influence of each parameter to the course of learning, and established the base of exploring the better values of parameters. This paper is also good for the researches on the effective solution to the problems of huge state spaces.