Diff-RNTraj: A Structure-aware Diffusion Model for Road Network-constrained Trajectory Generation

TL;DR

The paper introduces road network-constrained trajectory generation (RNTraj) and presents Diff-RNTraj, a diffusion-based framework that vectorizes hybrid road-segment and moving-ratio data into a continuous space, then decodes back to road-network trajectories. It addresses the challenge of generating on-road trajectories by combining a pre-trainedUTGraph-based road-segment representation with a RDCL-based denoising network and a spatial validity loss to enforce connectivity on the road graph. Key contributions include (i) a novel RNTraj vectorization and decoding pipeline, (ii) a structure-aware diffusion model for hybrid data, and (iii) extensive experiments on Porto and Chengdu datasets showing superior distributional fidelity and spatial validity, with strong performance in downstream trajectory prediction and robustness to limited data. The framework enables end-to-end, on-road trajectory generation with direct road-related information, reducing or eliminating the need for map-matching and enhancing applicability for traffic simulation, routing, and privacy-preserving data augmentation.

Abstract

Trajectory data is essential for various applications as it records the movement of vehicles. However, publicly available trajectory datasets remain limited in scale due to privacy concerns, which hinders the development of trajectory data mining and trajectory-based applications. To address this issue, some methods for generating synthetic trajectories have been proposed to expand the scale of the dataset. However, all existing methods generate trajectories in the geographical coordinate system, which poses two limitations for their utilization in practical applications: 1) the inability to ensure that the generated trajectories are constrained on the road. 2) the lack of road-related information. In this paper, we propose a new problem to meet the practical application need, \emph{i.e.}, road network-constrained trajectory (RNTraj) generation, which can directly generate trajectories on the road network with road-related information. RNTraj is a hybrid type of data, in which each point is represented by a discrete road segment and a continuous moving rate. To generate RNTraj, we design a diffusion model called Diff-RNTraj. This model can effectively handle the hybrid RNTraj using a continuous diffusion framework by incorporating a pre-training strategy to embed hybrid RNTraj into continuous representations. During the sampling stage, a RNTraj decoder is designed to map the continuous representation generated by the diffusion model back to the hybrid RNTraj format. Furthermore, Diff-RNTraj introduces a novel loss function to enhance the spatial validity of the generated trajectories. Extensive experiments conducted on two real-world trajectory datasets demonstrate the effectiveness of the proposed model.

Figures (14)

• Figure 1: Comparison between the two-stage solution and our end-to-end solution. Two-stage solution: first generate GPS points, then perform map matching process. Our solution: directly generate trajectories on the road end-to-end.
• Figure 2: A RNTraj is a point sequence formed by $\langle q_1,q_2,q_3,q_4 \rangle$. Each RNTraj point $q$ can be determined by road segment $e$ and moving ratio $r$, e.g., the road segment of RNTraj point $q_2$ is $e_4$, and the moving ratio is $r_2$. For $q_3$, its road segment is $e_4$, and the moving ratio is $r_3$.
• Figure 3: The architecture of Diff-RNTraj, which consists of a RNTraj vectorization module, a diffusion module, and a RNTraj decoder module.
• Figure 4: The architecture of the denoising network, which is stacked by $L$ residual dilation convolution layers (RDCLs).
• Figure 5: Hyperparameter analysis of embedding dimensional $D$ of the pre-trained road segment representation.

Theorems & Definitions (4)

• Definition 1: Trajectory
• Definition 2: Road Network
• Definition 3: Road Network-constrained Trajectory
• Definition 4: Road Network Constrained Trajectory Generation