Integrating State Representation Learning into Deep Reinforcement Learning