Relative Transitions, Not Absolute Destinations: A Transfer-and-Ground Framework for Target-Trajectory-Free Human Mobility Generation
The researchers developed Nomad, a transfer-and-ground framework that generates individual mobility trajectories for target cities without using any local trajectory data or derived statistics during training. The model learns transition priors over semantic displacement, geographic distance, and elapsed time from source city trajectories, then grounds these transitions onto a target city's POI map using a behavior graph and an exploration-return walk. Evaluated across 10 cities and 14 transfer scenarios, the framework reduced distributional fidelity errors by approximately 15% and improved downstream utility by about 3% compared to adaptation baselines.
Why it matters — It enables realistic human mobility simulation in data-scarce cities where trajectory data is completely unavailable but basic point-of-interest maps are accessible, bypassing the need for local training data or target-city model tuning.