2021-07 Reinforcement Learning with Prototypical Representations