LLMs for Human Mobility: Opportunities, Challenges, and Future Directions

TL;DR

This survey reviews how large language models are applied to human mobility tasks, highlighting opportunities, challenges, and future research directions in understanding and predicting human movement patterns.

Contribution

It provides the first comprehensive overview connecting human mobility tasks with LLM designs, synthesizing existing research and identifying open challenges.

Findings

LLMs effectively model semantic and contextual aspects of mobility.

Current approaches face challenges in reliability and privacy.

Future directions include grounded and privacy-aware LLM solutions.

Abstract

Human mobility studies how people move among meaningful places over time and how these movements aggregate into population-level patterns that shape accessibility, congestion, emissions, and public health. Large language models (LLMs) are increasingly used in this domain because many human mobility problems require reasoning about place and activity semantics, travelers' intentions and preferences, and diverse real-world constraints that are difficult to capture using coordinates and other purely numerical attributes. Despite rapid growth, the literature is still scattered, and there is no clear overview that connects human mobility tasks, challenges, and LLM designs in a consistent way. This survey therefore provides a comprehensive synthesis of LLM-based research on human mobility across five tasks, including travel itinerary planning, trajectory generation, mobility simulation, mobility prediction, and mobility semantics and understanding. For each task, we review representative work, connect core challenges to the specific roles of LLMs, and summarize typical LLM-based solution designs. We conclude with open challenges and research directions toward reliable, grounded and privacy-aware LLM-based approaches for human mobility.