DoesSelf-supervisedLearningReallyImprove ReinforcementLearningfromPixels?

Neural Information Processing Systems 

We investigate whether self-supervised learning (SSL) can improve online reinforcement learning (RL) from pixels.