Teleoperation is necessary for training humanoid robots, but reinforcement learning and simulation are still necessary, says ...