ControlTraj: Controllable Trajectory Generation with Topology-Constrained Diffusion Model

TL;DR

This work tackles the challenge of generating high-fidelity GPS trajectories while respecting road network topology to preserve realism and privacy. It introduces ControlTraj, a topology-constrained diffusion framework that combines a road-segment autoencoder (RoadMAE) for topology guidance with a geographic-attention UNet (GeoUNet) to steer diffusion-based trajectory synthesis. The approach enables controllable trajectory generation through conditional topology and trip attributes, and demonstrates strong fidelity and generalizability across three real-city datasets, with maintained utility in downstream traffic forecasting tasks. Zero-shot experiments further show that ControlTraj generalizes to unseen urban topologies, outperforming baselines that lack topology-aware guidance. Key methodological innovations include RoadMAE for robust road topology embeddings, GeoUNet with geo-attention to fuse topology and attributes into diffusion denoising, and a conditional diffusion objective that enables controllable trajectory generation. The empirical results show superior spatial-temporal fidelity, effective controllability to follow predefined routes, and practical data utility for traffic analysis, suggesting strong potential for privacy-preserving mobility studies and city-scale simulations.

Abstract

Generating trajectory data is among promising solutions to addressing privacy concerns, collection costs, and proprietary restrictions usually associated with human mobility analyses. However, existing trajectory generation methods are still in their infancy due to the inherent diversity and unpredictability of human activities, grappling with issues such as fidelity, flexibility, and generalizability. To overcome these obstacles, we propose ControlTraj, a Controllable Trajectory generation framework with the topology-constrained diffusion model. Distinct from prior approaches, ControlTraj utilizes a diffusion model to generate high-fidelity trajectories while integrating the structural constraints of road network topology to guide the geographical outcomes. Specifically, we develop a novel road segment autoencoder to extract fine-grained road segment embedding. The encoded features, along with trip attributes, are subsequently merged into the proposed geographic denoising UNet architecture, named GeoUNet, to synthesize geographic trajectories from white noise. Through experimentation across three real-world data settings, ControlTraj demonstrates its ability to produce human-directed, high-fidelity trajectory generation with adaptability to unexplored geographical contexts.

Figures (20)

• Figure 1: Comparison of trajectory generation models. Our controllable trajectory generator can provide high-fidelity trajectories and adhere to given conditions. Also, it adapts to new urban environments without retraining.
• Figure 2: The overview of the proposed ControlTraj framework. The pre-trained RoadMAE encodes road embedding based on the topology constraints of the road segments. Road embedding is then concatenated with trip attributes and merged into the diffusion model via geographic attention.
• Figure 3: The pipeline of RoadMAE.
• Figure 4: Visualization for different trajectory generation methods. Parameter size: DiffTraj 61.8 MB, ControlTraj 40.7 MB.
• Figure 5: Activity heatmap for periods within a day.