Most AI robots move like a buffering video: watch, freeze, think, act, repeat. A new MIT method called VLASH has them predict where they will be by the time they finish thinking, so they plan the next move while finishing the last one, smoothly enough to hold a real ping-pong rally with a human.
