To build general robotic agents that can operate in many environments, it is
often imperative for the robot to collect experience in the real world.
However, this is often not feasible due to safety, time, and hardware
restrictions. We thus propose leveraging the next best thing as rea