State-free Reinforcement Learning