https://zhengzhong-you.github.io/tags/trajectory-learning/Recent content in Trajectory Learning on Zhengzhong Ricky YouHugo -- 0.147.2enSun, 20 Sep 2026 00:00:00 +0000https://zhengzhong-you.github.io/papers/routing-foundation-models/Sun, 20 Sep 2026 00:00:00 +0000https://zhengzhong-you.github.io/papers/routing-foundation-models/<p>Accepted as a vision paper for presentation at the 2nd ACM SIGSPATIAL International Workshop on Urban Mobility Foundation Models (UMFM 2026). The acceptance notification was issued on September 20, 2026.</p> <p>The official program lists the paper in the Vision Papers session on November 3, 2026, in Riverside, California.</p> <ul> <li>Full paper: <a href="https://zhengzhong-you.github.io/papers/routing-foundation-models-umfm-2026.pdf">PDF</a> </li> <li>Workshop program: <a href="https://rsvp.withgoogle.com/events/sigspatial-2026-umfm-workshop" target="_blank" rel="noopener noreferrer">UMFM 2026</a> </li> </ul> <p>Commercial fleet trajectories record recurring operational decisions across vehicles, regions, and time. This vision paper proposes a research roadmap for using these trajectories to pretrain routing foundation models. The proposed studies compare spatiotemporal representations, separate the effects of data volume and coverage from model capacity and compute, and test transfer across regions and routing tasks. The roadmap also examines whether synthetic trajectories can transfer benefits to independent models while assessing disclosure risks.</p>