Reinforcement Learning and Approximate Dynamic