Siting Zhu

ROADGS-T achieves superior road surface mapping by dynamically adjusting surfel density based on local geometric complexity, revolutionizing how we perceive road features in autonomous driving.

Achieve photorealistic multi-traversal 3D reconstruction by disentangling illumination from material properties, even when the same scene is captured under drastically different lighting conditions.

Robots can now build and understand 3D scene graphs on the fly, thanks to a new system that incrementally maps semantics with Gaussian confidence.