Xinhao Liu

Papers

Total Citations

H-Index

About

Xinhao Liu is a researcher working at the intersection of autonomous systems, 3D spatial perception, and embodied artificial intelligence. His work spans LiDAR-based mapping, point cloud registration, and vision-driven urban navigation — areas critical to the advancement of self-driving vehicles and intelligent robotic agents. Liu's most notable contribution is **DeepMapping2**, a self-supervised framework that reformulates the notoriously difficult problem of large-scale LiDAR map optimization as the training of lightweight deep networks. By eliminating dependence on manual labels and extending the original DeepMapping approach to handle large-scale environments, this work has garnered significant attention in the robotics and autonomous driving communities, accumulating over 15 citations across its publications. The approach represents a meaningful step toward scalable, annotation-free 3D mapping pipelines. More recently, Liu has turned his attention to embodied navigation, introducing **CityWalker**, a system that learns urban navigation policies directly from web-scale video data. This work addresses the challenge of deploying agents in dynamic, map-free street environments — a frontier problem in embodied AI. Together, these contributions position Xinhao Liu as an emerging voice in perception-driven autonomy, bridging low-level 3D scene understanding with high-level agent behavior in real-world environments.