Editable Map-Conditioned Trajectory Generation for Human Mobility Simulation

2026-09-26

AI Summary

This study addresses the difficulty of data-driven generators in responding to map edits by proposing a map-conditioned autoregressive framework for human mobility simulation. Methodologically, road raster maps guide a decoder to generate trajectory grid tokens, integrating a grid-local vocabulary, a ResNet-50 visual prefix, and a Vision Transformer cross-attention architecture. This design enables local map editing without retraining and generalization to unseen maps. Experiments on smartphone data from Ishikawa Prefecture, Japan, demonstrate that interventions such as bridge removal instantly affect generated trajectories. Under correct map conditions, the generated trajectory density achieves a correlation of 0.38 with ground truth, and 60% of grids significantly approximate real-world distributions, validating the feasibility of editable mobility simulation.

Abstract

Geospatial simulation of infrastructure interventions requires mobility generators that respond directly to edited maps, yet many data-driven generators do not expose the map as an editable condition. We formulate this task as map-conditioned autoregressive generation of human mobility: a road raster conditions a decoder that emits nominal 31.25 m mesh-cell tokens at one-minute intervals. The mesh-local vocabulary supports held-out and locally edited maps without retraining or vocabulary changes. We instantiate a ResNet-50 visual-prefix configuration and a Vision Transformer (ViT) cross-attention configuration, trained from scratch on 87,400 smartphone-derived trajectories from 874 meshes in Ishikawa Prefecture, Japan; 219 meshes are held out. We evaluate map sensitivity by comparing correct-map and within-split shuffled-map generations with held-out real trajectories. On the 110-mesh test split, for the ResNet-50 configuration, correct-map generations are closer than shuffled-map generations on 60% of meshes under Hausdorff-based energy distance (p = 0.021), while DTW is directional but inconclusive (57%, p = 0.074); correlation with real density is 0.38 with the correct map versus 0.01 with shuffled maps. The ViT configuration shows weaker trajectory-level sensitivity and smaller density gains. An illustrative bridge-removal edit changes generated continuations without retraining. Together, these results support the feasibility of editable-map human-mobility simulation.

Problem

• human mobility simulation
• map-conditioned generation
• trajectory generation
• geospatial simulation
• editable map

Innovation

• Map-Conditioned Generation
• Autoregressive Trajectory Modeling
• Editable Maps
• Visual-Prefix Architecture
• Human Mobility Simulation