Reinforcement Learning for Optimal Feedback Control