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

2026-05-28

AI Summary

Existing trajectory generation methods struggle to explicitly model travel semantics and are constrained by fixed lengths and single-condition inputs, resulting in limited diversity and controllability. To address these limitations, this work proposes a hierarchical trajectory generation framework, HTP. The approach first introduces a trajectory-aware Residual Quantized Variational Autoencoder (RQ-VAE) to hierarchically abstract GPS trajectories into high-level mobility pattern tokens. It then extends the vocabulary of a large language model (LLM) and incorporates supervised fine-tuning to enable semantic-controllable and variable-length trajectory synthesis. This study presents the first application of a hierarchical generative paradigm to trajectory modeling, achieving an average improvement of 29.78% over the strongest baseline in generation quality across two real-world datasets, significantly enhancing realism, diversity, and conditional adaptability.

Abstract

Urban trajectories play a crucial role in modeling urban dynamics and supporting various smart city applications. However, privacy concerns restrict access to large-scale and high-quality trajectory datasets. Trajectory generation provides a promising alternative by synthesizing realistic data to mitigate privacy risks. However, existing methods fail to explicitly capture travel patterns and can only generate fixed-length trajectories under a single condition. To address these limitations, we propose \textbf{HTP}, which \textbf{H}ierarchically generates \textbf{T}ravel patterns first and then generates GPS \textbf{P}oints by using large language models (LLMs), rather than directly generating GPS points. We first design a trajectory-specific residual quantization variational autoencoder (RQ-VAE) that quantizes micro-level GPS trajectories into compact, macro-level travel pattern tokens in a coarse-to-fine manner. These tokens capture rich segment spatial irregularities, such as point density variations caused by traffic conditions. Then, we extend the LLM vocabulary with travel pattern tokens to align trajectory representations with the LLM input, and apply supervised fine-tuning (SFT) to align the LLM with the trajectory generation task, enabling generation of travel pattern sequences under various conditions. Extensive experiments on two real-world datasets show that HTP outperforms the strongest baseline by an average of 29.78\% in terms of generation quality. Our code is available at https://github.com/slzhou-xy/HTP.

Problem

• trajectory generation
• travel patterns
• GPS points
• privacy preservation
• urban dynamics

Innovation

• trajectory generation
• large language models
• travel patterns
• RQ-VAE
• semantic representation