UVTM: Universal Vehicle Trajectory Modeling with ST Feature Domain Generation

TL;DR

UVTM proposes a universal vehicle trajectory model that can be trained once and adapted to multiple trajectory tasks, addressing incomplete and sparse feature scenarios. It achieves this through a three-domain feature masking framework and a pre-training regime that reconstructs dense trajectories from sparse inputs, reinforced by contrastive learning. A hierarchical trajectory encoder with dual-layer positions and a specialized output layer enables task-specific predictions without extensive retraining. Experiments on three real-world datasets across four tasks show UVTM’s robust performance, cross-task transfer, and efficiency advantages over task-specific and existing universal models. This work substantially reduces the need for multiple specialized models while delivering reliable ITS analytics under challenging data conditions.

Abstract

Vehicle movement is frequently captured in the form of GPS trajectories, i.e., sequences of timestamped GPS locations. Such data is widely used for various tasks such as travel-time estimation, trajectory recovery, and trajectory prediction. A universal vehicle trajectory model could be applied to different tasks, removing the need to maintain multiple specialized models, thereby reducing computational and storage costs. However, creating such a model is challenging when the integrity of trajectory features is compromised, i.e., in scenarios where only partial features are available or the trajectories are sparse. To address these challenges, we propose the Universal Vehicle Trajectory Model (UVTM), which can effectively adapt to different tasks without excessive retraining. UVTM incorporates two specialized designs. First, it divides trajectory features into three distinct domains. Each domain can be masked and generated independently to accommodate tasks with only partially available features. Second, UVTM is pre-trained by reconstructing dense, feature-complete trajectories from sparse, feature-incomplete counterparts, enabling strong performance even when the integrity of trajectory features is compromised. Experiments involving four representative trajectory-related tasks on three real-world vehicle trajectory datasets provide insight into the performance of UVTM and offer evidence that it is capable of meeting its objectives.

Figures (9)

• Figure 1: Overview of a typical flow of trajectory modeling, and scenarios involving incomplete or sparse trajectories that pose challenges for trajectory modeling.
• Figure 2: A vehicle trajectory and its map-matched counterpart.
• Figure 3: Two core components of the proposed UVTM: tuple of feature domains and auto-regressive generation of the tuples.
• Figure 4: Pipeline of the feature domain embedding layer. Continuous features are encoded with learnable Fourier features, while discrete features use index-fetching embedding modules.
• Figure 5: Architecture of the hierarchical attention in the trajectory encoder.

Theorems & Definitions (5)

• Definition 1: Road Network
• Definition 2: GPS Trajectory
• Definition 3: Sampling Interval
• Definition 4: Map-matched Trajectory
• Example 1