How world action models let robots forecast what will happen next as video, points or goal images, and turn that imagined future into actions, from robot-tailored video pretraining and learning from unlabeled footage to long-horizon planning and physically faithful forecasts.