Reinforcement learning roadmap