Hang Dai 戴航

I am an undergraduate student majoring in Information and Computing Science at Peking University, with expected graduation in June 2027.

I want robots to learn from 4D worlds that are physically correct, not just visually realistic. A scene can look convincing while its objects float without support, pass through each other, or move with the wrong dynamics, and a robot trained in such a world learns the wrong lessons. I therefore treat geometry, articulation, and dynamics as first-class targets whenever I reconstruct or generate the world.

My work builds such worlds from two sources. From the real world, I reconstruct simulation-ready articulated objects from a single free-moving monocular RGB-D video (FreeArtGS, CVPR 2026), and build physics-aware digital twins whose visual and dynamic alignment enables zero-shot real-to-sim-to-real transfer of manipulation policies (TwinAligner). From imagination, I build agentic systems that generate minute-long 4D scenes, spanning humanoid robots, characters, and physical mechanisms, by planning a 3D storyboard of verified key poses and authoring motion between them, with substantially fewer physical violations than frontier coding agents (DyCraft).

Going forward, I aim to make reconstructed and generated worlds a scalable source of training data, simulation environments, and evaluation for robot learning.

I strive to balance rigorous academic coursework, engaging research projects, and meaningful social work. Always eager to learn and collaborate, I welcome discussions on innovative technologies and interdisciplinary initiatives.