Reinforcement Learning in Structured and Partially Observable Environments