A joint travel mode and departure time choice model in dynamic multimodal transportation networks based on deep reinforcement learning
Ziyuan Gu, Yukai Wang, Wei Ma, and Zhiyuan Liu
Multimodal Transportation, 3(3), 100137
Publications, submissions & research development
Research record
Published work, manuscripts under review, and ongoing projects are listed separately below. Current research status is updated as of 22 September 2026.
Journal article
Conference contributions
ProSafeAV: Enhancing Proactive Safety Performance of AVs using World Model-Based Reinforcement Learning and Extreme Value Theory
Yukai Wang, Tiantian Chen, and Sikai Chen
TRB Annual Meeting 2025
A Synergistic Approach to Real-Time Crash Risk Estimation at Signalized Intersections: Integrating Learning-Based Anomaly Detection with Extreme Value Theory
Yukai Wang, Tiantian Chen, and Kitae Jang
Conference in Emerging Technologies in Transportation Systems (TRC-30), September 2–4, 2024
Manuscripts under review
Crash risk estimation using auxiliary bivariate anomaly scoring and non-stationary bivariate peak over threshold modeling
Traffic-conflict risk estimation with anomaly scoring and extreme value theory; revised manuscript.
Proactive Risk Modeling for Autonomous Driving with World Models and Extreme Value Theory
ProSafeAV: linking imagined conflict severity to policy learning; revised manuscript.
Related research →Where Closed-Loop Traffic World Models Break and When Correction Pays
Paired diagnostics for the source and timing of closed-loop rollout errors.
Related research →Missing-History Robustness Is Not a Single Number
Separating input-level missing-history tests from latent-pathway diagnostics.
Related research →Ongoing research
These active projects form part of my broader research programme. They are listed separately from the manuscripts currently under review.
World models from drone trajectories
Develop graph-conditioned recurrent state-space models with neural ODE motion decoding; evaluate multimodal prediction and observation-grounded versus self-conditioned rollout on drone trajectory data.
From predictive validity to decision utility
Build recording-disjoint, future-free candidate-ranking audits with fixed candidate banks, analytic baselines, leakage checks, rank fidelity, and reference-relative regret.
Safety assessment from forecasted trajectories
Develop leakage-controlled pipelines from predicted trajectories to traffic-conflict indicators, event-level diagnostics, and tail-risk analysis.
Behavioral validity and cross-site generalization
Investigate layered behavioral validity in calibrated microsimulation and cross-site evaluation of graph-based and autoregressive traffic models.
For bibliographic updates, see Google Scholar.