UC Berkeley, Meta model predicts egocentric video from motion

The model currently handles a limited range of actions, lacks the ability for long-term planning or complex task execution.