Found this going down a robot-control rabbit hole and it stuck with me. The belt keeps moving, so the target never sits still, which is the kind of thing that usually makes a robot lag or fumble. This one keeps pace by predicting where the scene is about to go and acting on that, then correcting on every new camera frame, instead of only reacting to the current instant. It is a video-action model called LingBot-VA 2.0. The clip is 1x with no cuts, so nothing is sped up. I will drop the source and the honest limits in a comment instead of overselling it here. Curious what people here make of it.
[link] [comments]