Guiding LLM-Based Human Mobility Simulation with Mobility Measures from Shared Data

2026-02-20 · Source: cs.MA updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, extended

Summary

M2LSimu is a novel framework designed to enhance the realism of large language model (LLM)-based human mobility simulations by integrating population-level mobility measures from shared data. Traditional LLM approaches generate individual trajectories independently, often failing to capture collective behaviors like scaling laws. M2LSimu addresses this by using a multi-prompt adjustment framework that refines individual-level prompts based on coarse-grained mobility measures, such as radius of gyration, travel distance, and visitation frequency. This iterative process, formulated as a Markov Decision Process and optimized with Monte Carlo Tree Search, enables fine-grained individual adaptation while satisfying multiple population-level mobility objectives under a limited computational budget. Experiments on Beijing and New York City datasets demonstrate that M2LSimu significantly outperforms existing LLM-based methods, with improvements ranging from 11.29% to 64.08% across various metrics, even when guided by statistical summaries without trajectory-level data.

Key takeaway

For research scientists developing large-scale human mobility simulations, M2LSimu offers a robust method to overcome the limitations of independent LLM agent generation. You should consider integrating population-level mobility measures as guidance for prompt adjustment, even with coarse-grained or statistical data, to achieve more realistic collective behaviors. This approach not only improves simulation accuracy but also allows for efficient scaling of optimized prompts across larger populations, making it a practical solution for urban science, epidemiology, and transportation analysis.

Key insights

Guiding LLM-based mobility simulations with population-level measures improves collective behavior realism and efficiency.

Principles

• Collective behaviors emerge from coordinated individual actions.
• Coarse-grained data can effectively guide fine-grained adjustments.
• Multi-objective optimization balances diverse simulation goals.

Method

M2LSimu formulates prompt adjustment as a Markov Decision Process, using Monte Carlo Tree Search to iteratively refine individual LLM prompts based on population-level mobility measures derived from shared data, ensuring collective behavior realism.

In practice

• Use mobility measures to refine LLM prompts for realistic simulations.
• Apply MCTS for multi-objective prompt optimization.
• Generalize optimized prompts from subsets to larger populations.

Topics

• Human Mobility Simulation
• Large Language Models
• Prompt Optimization
• Multi-objective Optimization
• Monte Carlo Tree Search

Best for: Research Scientist, AI Researcher, AI Scientist, Prompt Engineer

Related on AIssential

• MobEvolve: An Agentic Self-Evolving Heuristic System for Interpretable Human Mobility Generation — MobEvolve uses an LLM agent to self-evolve heuristic rules for human mobility generation, achieving superior performance and interpretability.
• Towards Real-world Human Behavior Simulation: Benchmarking Large Language Models on Long-horizon, Cross-scenario, Heterogeneous Behavior Traces — Real-world human behavior simulation requires long-horizon, cross-scenario data, which current LLMs struggle to model accurately.
• Large Language Models in Transportation Systems Management and Operations: From Text Reasoning to Multi-modal Decision Support — LLMs and MM-LLMs offer a new mechanism for integrating diverse transportation data into operator-facing decision support systems.
• DIVE: Scaling Diversity in Agentic Task Synthesis for Generalizable Tool Use — Scaling diversity in agentic task synthesis through evidence-first execution significantly improves LLM generalization for tool use.
• Weight Patching: Toward Source-Level Mechanistic Localization in LLMs — Weight Patching identifies where LLM capabilities are encoded in parameters, distinguishing source carriers from aggregation modules.
• Beyond Static Responses: Multi-Agent LLM Systems as a New Paradigm for Social Science Research — A six-tier framework maps LLM agent evolution from static tools to complex adaptive systems for social science research.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.MA updates on arXiv.org.