“Danny Drace asserts that current generative video models are not yet useful for robotics simulation because they fail to accurately model physics, especially when a robot is in the scene, leading to unrealistic outputs.”