Learning Generalized and Flexible Trajectory Models from Omni-Semantic Supervision

TL;DR

OmniTraj addresses the challenge of scalable, flexible trajectory retrieval by introducing four modality-specific encoders (trajectory, topology, road, region) whose embeddings are aligned in a shared latent space. It employs a simple, modular architecture and a bidirectional InfoNCE contrastive objective to enable queries using any single modality, any combination, or a learned fusion of modalities. The experimental results on Chengdu and Xi'an show that OmniTraj outperforms classical similarity measures and prior self-supervised methods, with notable gains in condition-based retrieval and multi-modality querying, and demonstrate transferability and scalability to city-scale data. The work offers practical implications for intelligent transportation and spatial analytics by enabling semantically rich, efficient, and flexible trajectory retrieval at large scale.

Abstract

The widespread adoption of mobile devices and data collection technologies has led to an exponential increase in trajectory data, presenting significant challenges in spatio-temporal data mining, particularly for efficient and accurate trajectory retrieval. However, existing methods for trajectory retrieval face notable limitations, including inefficiencies in large-scale data, lack of support for condition-based queries, and reliance on trajectory similarity measures. To address the above challenges, we propose OmniTraj, a generalized and flexible omni-semantic trajectory retrieval framework that integrates four complementary modalities or semantics -- raw trajectories, topology, road segments, and regions -- into a unified system. Unlike traditional approaches that are limited to computing and processing trajectories as a single modality, OmniTraj designs dedicated encoders for each modality, which are embedded and fused into a shared representation space. This design enables OmniTraj to support accurate and flexible queries based on any individual modality or combination thereof, overcoming the rigidity of traditional similarity-based methods. Extensive experiments on two real-world datasets demonstrate the effectiveness of OmniTraj in handling large-scale data, providing flexible, multi-modality queries, and supporting downstream tasks and applications.

Figures (8)

• Figure 1: Comparison of trajectory retrieval methods. Our method can retrieve trajectories according to their different modalities and provides flexible retrieval solutions.
• Figure 2: Overview of the proposed OmniTraj framework, which contains four individual modal encoders. Each encoder captures a diverse semantic representation of a trajectory and projects it into a shared space.
• Figure 3: Model structure and parameters and setting.
• Figure 4: Model transferability and scalability analysis.
• Figure 5: Use OmniTraj as a semantic representation to guide the model for trajectory generation.