Reinforcement Learning with Prototypical Representations

              Denis Yarats 1 2 Rob Fergus 1 Alessandro Lazaric 2 Lerrel Pinto 1

              Abstract               from rewards alone is sample inefficient and leads to poor
    Learning effective representations in image-based    performance. Prior work (Srinivas et al., 2020; Yarats et al.,
    environments is crucial for sample efficient Rein-    2019) addresses this problem by leveraging self-supervised
    forcement Learning (RL). ...