When Plausible Is Not Realistic: Evaluating Human Mobility in LLM-Based Urban Simulation

2026-06-11

AI Summary

This study addresses the lack of systematic evaluation of the realism of human mobility generated by large language model (LLM)-driven urban simulators. The authors propose a multidimensional validation framework that leverages real-world urban data to assess the fidelity of LLM-based agents across several dimensions: mobility patterns, temporal rhythms, network motifs, semantic activity transitions, and behavioral profiles. Their analysis reveals, for the first time, a significant gap between narrative plausibility and empirical realism in simulated behaviors, highlighting the necessity of explicitly initializing behavioral profiles to enhance both diversity and fidelity. Through mobility law analysis, spatiotemporal network modeling, and semantic sequence mining, experiments on datasets from the Paris metropolitan area and Shanghai demonstrate that current simulators struggle to reproduce key empirical features such as trip distances, origin–destination flows, dwell durations, and transition dynamics. The study also releases an open-source, reproducible evaluation infrastructure.

Abstract

LLM-based generative agents are increasingly used in urban simulators, yet it remains unclear whether they reproduce empirically realistic human mobility patterns or merely generate plausible mobility narratives. We introduce a validation framework for evaluating the mobility of generative agents of LLM-based urban simulators against real-world mobility data. For this, we use mobility laws, temporal rhythms, network motifs, semantic activity transitions, and behavioral mobility profiles. Using datasets from the Greater Paris region and Shanghai, we evaluate AgentSociety and CitySim across multiple dimensions of mobility realism. Our analysis reveals a substantial gap between narrative plausibility and empirical mobility realism. Although the simulators capture some high-level semantic activity distributions, they struggle to reproduce core spatial and temporal constraints, including realistic trip-length distributions, origin-destination flows, dwell times, and transition dynamics. We further observe that realistic mobility diversity is unstable across default prompting configurations and may require explicit profile-aware initialization. To support reproducible evaluation, we also contribute scalable and open LLM-driven infrastructure for regional-scale map generation, observability-enhanced simulation, mobility-metric computation, and traffic simulation. Our findings highlight the need for rigorous empirical validation of LLM-based urban simulators and provide practical tools for building more realistic and reproducible urban simulation systems.

Problem

• human mobility
• LLM-based urban simulation
• mobility realism
• generative agents
• empirical validation

Innovation

• LLM-based urban simulation
• human mobility validation
• generative agents
• mobility realism
• simulation reproducibility