Discuss the challenges of transferring a reinforcement learning policy trained in simulation to a physical robot for Berkshire Grey.