Resnet as representation for Reinforcement Learning (RRL) is a simple yet effective approach for training behaviors directly from visual inputs. We demonstrate that features learned by standard image