From GPS Points to Travel Patterns: Flexible and Semantic Trajectory Generation with LLMs

2026-05-28 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Robotics & Autonomous Systems · Depth: Advanced, quick

Summary

HTP, a novel model, addresses limitations in urban trajectory generation by hierarchically synthesizing travel patterns before generating GPS points using large language models. Existing methods struggle with privacy concerns and cannot explicitly capture diverse travel patterns or generate flexible-length trajectories under varied conditions. HTP first employs a trajectory-specific residual quantization variational autoencoder (RQ-VAE) to convert micro-level GPS data into macro-level travel pattern tokens, capturing segment spatial irregularities. It then extends an LLM's vocabulary with these tokens and applies supervised fine-tuning to enable conditional generation of travel pattern sequences. Experiments on two real-world datasets demonstrate HTP's effectiveness, outperforming the strongest baseline by an average of 29.78% in generation quality.

Key takeaway

For AI Scientists developing smart city applications, HTP offers a robust solution for generating high-quality, privacy-preserving urban trajectory data. If you are constrained by limited access to real-world datasets due to privacy concerns, consider implementing HTP's hierarchical LLM-based approach. This method allows for flexible, conditional trajectory synthesis, potentially accelerating research and development in urban dynamics modeling without compromising sensitive information.

Key insights

HTP uses LLMs and a hierarchical RQ-VAE to generate realistic urban trajectories by first modeling travel patterns.

Principles

• Hierarchical generation improves trajectory realism.
• Quantizing GPS data into tokens aligns with LLM inputs.
• LLMs can model complex urban travel patterns.

Method

HTP first quantizes micro-level GPS trajectories into macro-level travel pattern tokens using an RQ-VAE. It then extends an LLM's vocabulary with these tokens and applies supervised fine-tuning for conditional sequence generation.

In practice

• Synthesize realistic urban trajectory datasets.
• Mitigate privacy risks in smart city applications.
• Generate varied travel patterns under conditions.

Topics

• Urban Trajectory Generation
• Large Language Models
• Residual Quantization VAE
• Smart City Applications
• Data Privacy
• Supervised Fine-tuning

Code references

• slzhou-xy/HTP

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• Broadening Access to Transportation Safety Data with Generative AI: A Schema-Grounded Framework for Spatial Natural Language Queries — A schema-grounded natural language interface broadens access to transportation safety data while ensuring reproducible, auditable analysis.
• Inroads to personalized AI trip planning — Combining LLMs with formal solvers enables robust, verifiable solutions for complex, multi-constraint planning problems.
• A Dual Perspective on Synthetic Trajectory Generators: Utility Framework and Privacy Vulnerabilities — Evaluating synthetic trajectory generators requires a dual focus on utility and privacy, especially against adversarial attacks.
• Teaching AI to read a map — Explicitly teaching spatial reasoning to MLLMs through synthetic data significantly improves their navigation capabilities.
• AWS Generative AI Model Agility Solution: A comprehensive guide to migrating LLMs for generative AI production — A structured framework for LLM migration and optimization ensures continuous performance improvement and operational stability.
• TransitLM: A Large-Scale Dataset and Benchmark for Map-Free Transit Route Generation — End-to-end, map-free transit route generation is feasible by training LLMs on large-scale trajectory data, enabling implicit spatial grounding.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.