Xiangyu Li

Papers

Total Citations

H-Index

About

Xiangyu Li is a researcher specializing in human trajectory prediction and intelligent motion forecasting, with particular expertise in deep learning architectures applied to autonomous systems and computer vision. Li's work addresses critical challenges in predicting pedestrian and human movement patterns — capabilities foundational to autonomous driving, robot navigation, and intelligent surveillance technologies. Li's most notable contribution, "MRGTraj" (2023), introduced a pioneering non-autoregressive framework for trajectory prediction, directly tackling a fundamental limitation of conventional RNN- and Transformer-based models: the accumulation of errors inherent in sequential, autoregressive generation. This innovative approach has already garnered 23 citations, signaling strong community recognition shortly after publication. Earlier work on "PECGAN" (2021) demonstrated Li's sustained engagement with generative modeling techniques, leveraging Generative Adversarial Networks conditioned on predicted endpoints to improve trajectory forecasting accuracy — contributing to the growing intersection of GANs and motion prediction research. Across these contributions, Li has demonstrated a consistent drive to rethink architectural assumptions in sequential modeling, pushing toward more robust and efficient prediction systems. For students and researchers exploring autonomous navigation or pedestrian behavior modeling, Li's work offers valuable methodological insights into both generative and non-autoregressive approaches to motion forecasting.