Jian Sun, Ruoxi Kong, Xiaocong Zhao, Haoyang Liang, Xiyan Jiang, and Yiru Liu

SSRN Preprint, 2026

Paper Paper

• A physics-informed interaction modeling framework makes ego-centric interaction semantics explicit and quantifiable for warning targeting in dense multi-risk traffic.

• The interaction-aware HMI improves critical-agent identification and yields faster attention shifts with far fewer collisions in driver-in-the-loop virtual reality experiments.

Xiyan Jiang, Xiaocong Zhao, Yiru Liu, Peng Hang, and Jian Sun

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC)

Paper Paper

• We enumerate all dynamically feasible multi-agent interaction modes in a scene by decomposing negotiations at conflict points and solving graph-based spatiotemporal anchors.

• A controllable diffusion model then generates realistic full-scene trajectories, including low-probability but physically and socially plausible counterfactuals for AV testing.

Xiyan Jiang, Xiaocong Zhao*, Yiru Liu, Zirui Li, Peng Hang, Lu Xiong, and Jian Sun

GitHub Code Paper Paper

• A unified naturalistic driving dataset with dense interactions is built by integrating and organizing multiple public datasets, providing a clear and accessible resource for studying complex traffic behaviors.

• A formal, quantitative definition of driving interactions, paired with an open-source toolkit, enables effective analysis and application across autonomous driving tasks.