Learning Invariant Representations for Reinforcement Learning without Reconstruction - View it on GitHub