LLM-Based Crowd Simulation Using Aggregate Mobility Statistics
August 21, 2026
This method fine-tunes language model agents to match origin-to-destination flows and zone-level device counts in pedestrian simulations. It utilizes iterative proportional fitting and low-rank adapters to ensure simulated populations reproduce observed aggregate mobility patterns without individual trajectory data.
HOW THIS AFFECTS YOU
●
builderYou can generate privacy-compliant crowd simulations using only aggregate mobility data and fine-tuned LLMs.