Ying Wang, Oumayma Bounou, Yann LeCun, and Mengye Ren have a paper on AdaJEPA, a latent world model that updates during model predictive control instead of staying frozen after training. The retained page says the system executes an action chunk, uses the observed next-state transition as a self-supervised update signal, and replans with the changed model. The visible claim is better goal-reaching under visual and dynamics shifts, with as little as one gradient step per replanning step.