Relative Transitions, Not Absolute Destinations: A Transfer-and-Ground Framework for Target-Trajectory-Free Human Mobility Generation

2026-10-01

AI Summary

This study addresses the challenge of cross-city transferability in trajectory generation when target-city data is scarce, as conventional generators relying on absolute destinations struggle to generalize. To overcome this limitation, this work proposes Nomad, a framework that decouples mobility behavior from spatial layouts. Its core innovation lies in replacing absolute destination prediction with context-conditioned relative transitions, thereby achieving representation-level transferability. Specifically, Nomad learns relative transition priors from source-city trajectories and instantiates generation on the target city’s POI map by integrating a history-conditioned flow matching model, behavior graph construction, and an explore-return random walk algorithm. Experiments across ten cities demonstrate that Nomad reduces distribution fidelity errors by approximately 15% and downstream utility errors by roughly 3%, significantly outperforming baseline methods.

Abstract

Individual mobility trajectories support urban analysis and location-based services, yet most trajectory generators require observations from their deployment city. This assumption excludes precisely the cities where trajectories are unavailable even though points of interest (POIs) and their attributes can be obtained from public maps. We study target-trajectory-free generation: learning from POIs and trajectories in source cities while utilizing only POI coordinates and categories in a target city, with no target trajectory or trajectory-derived statistic available for training, model selection, or generation. Existing trajectory generators typically predict absolute destinations, entangling reusable movement behavior with city-specific POI identities and spatial layouts. Our core insight is to replace this city-bound output with context-conditioned relative transitions. We propose Nomad, a transfer-and-ground framework that separates learning how people move from determining where those movements are realized. Specifically, a history-conditioned flow-matching model learns from source trajectories a transition prior over semantic displacement between POI contexts, geographic displacement, and elapsed time; at inference, a behavior graph and an exploration--return walk ground sampled transitions onto the target POI map. This factorization enables a direct test of representation level transferability without assuming invariance of the full mobility distribution. Extensive experiments across ten cities and 14 transfers show that Nomad outperforms adaptation baselines in trajectory fidelity and downstream utility, lowering the average error over the best baseline of each metric by about 15% in distributional fidelity and about 3% in downstream utility.

Problem

• human mobility generation
• target-trajectory-free
• cross-city transfer
• point of interest
• trajectory prediction

Innovation

• target-trajectory-free generation
• transfer-and-ground framework
• relative transitions
• flow-matching model
• behavior graph