Learning robot skills from human demonstration videos by finding common action representations across different embodiments.