Dunqiang Liu

Papers

Total Citations

H-Index

About

Dunqiang Liu is a leading researcher in robotics and computer vision, specializing in LiDAR-based localization and language-driven spatial understanding. His work addresses critical bottlenecks in autonomous systems, particularly the trade-off between accuracy and computational efficiency. Liu’s landmark paper, “LightLoc,” revolutionizes outdoor LiDAR localization by slashing training times from days to near-instantaneous speeds—a breakthrough that makes scene coordinate regression practical for time-sensitive applications like autonomous driving. With over 3 citations in its first year, this work is already shaping real-time localization pipelines. In parallel, Liu tackles the challenge of language-based localization in “Text to Point Cloud Localization with Multi-Level Negative Contrastive Learning,” introducing novel contrastive learning techniques to resolve the inherent ambiguity in matching textual descriptions to 3D point clouds. By leveraging multi-level negative sampling, his method significantly improves the precision of spatial reasoning from natural language commands. Liu’s contributions are foundational for next-generation autonomous systems that require rapid adaptation to new environments and intuitive human-robot interaction. His research continues to push the boundaries of efficient, scalable localization—a cornerstone for safe and intelligent navigation in dynamic, unstructured worlds.