Yuwen Xiong

Papers

Total Citations

147

H-Index

About

Yuwen Xiong is a researcher specializing in autonomous driving perception, deep learning, and world modeling for robotic systems. His work sits at the intersection of computer vision and self-driving technology, where he has made meaningful contributions to some of the field's most challenging problems. Xiong's most recognized contribution is his work on **scene flow estimation**, presented in "Deep Rigid Instance Scene Flow" (2019), which has accumulated over 130 citations. This research introduced a novel approach to understanding how a scene dynamically changes by leveraging deep learning alongside strong geometric priors — decomposing scene motion into the robot's own movement and the independent 3D motion of surrounding actors. This framework proved particularly impactful for real-world autonomous driving applications, where accurate motion estimation is safety-critical. More recently, Xiong has pushed into the frontier of **unsupervised world models** for autonomous driving through "Copilot4D" (2023), which applies discrete diffusion models to help autonomous agents learn how environments evolve over time — an ambitious effort to bring the scaling successes of language models to robotic perception. Together, these works demonstrate a research trajectory focused on enabling machines to perceive, predict, and reason about dynamic environments — foundational capabilities for the next generation of intelligent autonomous systems.