Distributional reinforcement learning