AI Researchers from Stanford Propose ‘EquivAct’: A Breakthrough in Robot Learning for Generalizing Tasks Across Different Scales and Orientations