Shenghai Yuan

Real-world robots can now navigate complex environments with human-level instructions, thanks to a new system that combines efficient perception with high-level reasoning, all while running in real-time on limited hardware.

LLMs in embodied environments get a massive boost from structured rules, with rule retrieval alone contributing +14.9 pp to single-trial success.

Skip expensive manual annotation: this method extracts accurate 3D UAV trajectories and classifications directly from readily available internet videos.

Forget predefined areas of interest: this multi-agent exploration framework uses Gaussian belief mapping to adaptively balance scientific discovery and safety in hazardous off-world environments.

Panoramic depth perception and differentiable physics unlock surprisingly robust collision avoidance, even generalizing to unseen simulation environments.

Finally, a 14B video generation model that runs in real-time (19.5 FPS on a single H100) *without* relying on common acceleration tricks or anti-drift hacks.

Legged robots can now tiptoe around your expensive gadgets, thanks to a new RL framework that combines semantic understanding with low-level control to avoid stepping on designated objects.