We propose RoboScape, a unified physics-informed embodied world model that jointly learns video generation and physics knowledge within an integrated framework.