World models for traffic, beyond a single vehicle.

Yukai Wang 王于凯

Traffic world models · Multi-agent learning · Safety

I am a PhD Candidate at the Cho Chun Shik Graduate School of Mobility, KAIST, advised by Prof. Tiantian Chen, in the Human Factors Centered Transport Safety Laboratory.

My research asks how world models can serve transportation beyond the control of a single autonomous vehicle. I use drone-recorded trajectories to study the interactions of road users across an entire traffic site, with the aim of supporting traffic understanding, prospective safety assessment, and the evaluation of digital twins.

This direction grew from my work on traffic conflicts, anomaly detection, and extreme value theory. I wanted to move from measuring an interaction after it happened to understanding how it might unfold. That led to a question I keep returning to: when does a good trajectory predictor become a useful world model? My current studies test observation dependence, closed-loop error propagation, and whether better forecasts lead to better decisions.

Previously, I completed my master’s studies at Southeast University (2020–2023), advised by Prof. Zhiyuan Liu, and my bachelor’s studies at Southwest Jiaotong University (2016–2020), in Mao Yisheng Honors College.

This site brings together my published work, manuscripts under review, and ongoing research. Read the research story or view current manuscripts.