Physics
Moving beyond describing and generating worlds to predicting the consequences of action under real constraints.
From a world that looks right to one that behaves right
Representation asks what is here. Synthesis asks what else may be here and how it may change. Physics asks the harder question: what happens when an agent acts on the world?
Work in progress
This research direction is taking shape. We will add the research narrative and project links here when the results are ready.
The research question
The physical world does not obey visual plausibility alone. Mass, contact, friction, joints, deformation, and reaction to action introduce causal constraints on a world state. We are exploring how to connect disentangled 3D/4D representations and controllable dynamics with physical state, causal prediction, and interaction policies.
Our goal: not only to render a world, but to probe it, predict it, and act within it—so that every interaction becomes evidence for the next round of representation learning.