Reinforcement Learning with Soft State Aggregation