Robots learn better when they think about how things move over time, not by redrawing every pixel of a video.
Robots usually think in words and pictures, but their hands need exact motions, so there is a gap between understanding and doing.