OpenNovelty.org

TrajFlow: Nation-wide Pseudo GPS Trajectory Generation with Flow Matching Models

ICLR 2026 Conference SubmissionAnonymous Authors
OpenReview Score: 6.5Download Report PDF
Flow matchingHuman TrajectoryGenerative modelingHuman mobility
Abstract:

The importance of mobile phone GPS trajectory data is widely recognized across many fields, yet the use of real data is often hindered by privacy concerns, limited accessibility, and high acquisition costs. As a result, generating pseudo–GPS trajectory data has become an active area of research. Recent diffusion-based approaches have achieved strong fidelity but remain limited in spatial scale (small urban areas), transportation-mode diversity, and efficiency (requiring numerous sampling steps). To address these challenges, we introduce TrajFlow, the first flow-matching–based generative model for GPS trajectory generation. TrajFlow leverages the flow-matching paradigm to improve robustness across multiple geospatial scales and incorporates a trajectory harmonization & reconstruction strategy to jointly address scalability, diversity, and efficiency. Using a nationwide mobile phone GPS dataset with millions of trajectories across Japan, we show that TrajFlow consistently outperforms diffusion-based and deep generative baselines at urban, metropolitan, and nationwide levels. As the first nationwide, multi-scale GPS trajectory generation model, TrajFlow demonstrates strong potential to support inter-region urban planning, traffic management, and disaster response, thereby advancing the resilience and intelligence of future mobility systems.

Disclaimer
This report is AI-GENERATED using Large Language Models and WisPaper (A scholar search engine). It analyzes academic papers' tasks and contributions against retrieved prior work. While this system identifies POTENTIAL overlaps and novel directions, ITS COVERAGE IS NOT EXHAUSTIVE AND JUDGMENTS ARE APPROXIMATE. These results are intended to assist human reviewers and SHOULD NOT be relied upon as a definitive verdict on novelty.
NOTE that some papers exist in multiple, slightly different versions (e.g., with different titles or URLs). The system may retrieve several versions of the same underlying work. The current automated pipeline does not reliably align or distinguish these cases, so human reviewers will need to disambiguate them manually.
If you have any questions, please contact: mingzhang23@m.fudan.edu.cn

Overview

Overall Novelty Assessment

TrajFlow introduces the first flow-matching-based generative model for GPS trajectory synthesis, positioning itself within the 'Flow Matching and Diffusion Models' leaf of the taxonomy. This leaf contains only two papers, indicating a relatively sparse but emerging research direction. The paper's core contribution—applying flow matching to trajectory generation—represents a methodological shift from diffusion-based approaches (its sole sibling, Geo-lucid Diffusion) toward more efficient probabilistic frameworks. The taxonomy reveals this is a nascent area within the broader 'Deep Learning-Based Trajectory Generation' branch, which itself competes with privacy-preserving, rule-based, and domain-specific synthesis methods.

The taxonomy structure shows TrajFlow sits adjacent to several related directions. The 'Individual-Based Mobility Generation' leaf (three papers) focuses on personalized trajectory modeling, while 'High-Fidelity Synthetic Dataset Creation' (one paper) emphasizes benchmark dataset construction. These neighboring leaves share the goal of realistic synthesis but differ in conditioning strategies and scale. The 'Privacy-Preserving Trajectory Synthesis' branch (four papers across two leaves) represents an orthogonal concern—formal privacy guarantees—that TrajFlow does not explicitly address. The taxonomy's scope notes clarify that flow matching methods belong specifically to generative modeling, excluding rule-based approaches in the 'Road Network-Based Synthetic Trajectory Generation' leaf.

Among 26 candidates examined, no papers clearly refute TrajFlow's three main contributions. The first contribution (flow-matching framework) examined 10 candidates with zero refutations, suggesting novelty in applying this specific generative paradigm to GPS trajectories. The second contribution (unified harmonization and OD-conditioned normalization) examined 6 candidates, again with no refutations, indicating the integration strategy may be distinctive. The third contribution (nationwide multi-scale modeling) examined 10 candidates without refutation, though the limited search scope means prior work at similar geographic scales could exist beyond the top-26 semantic matches. The statistics reflect a focused literature search rather than exhaustive coverage.

Based on the limited search scope of 26 candidates, TrajFlow appears to occupy a relatively novel position within an emerging subfield. The sparse 'Flow Matching and Diffusion Models' leaf and absence of refutable prior work suggest methodological distinctiveness, though the analysis cannot rule out relevant work outside the top-K semantic neighborhood. The nationwide scale claim warrants particular caution, as geographic coverage may not be well-captured by semantic search alone.

Taxonomy

Core-task Taxonomy Papers
29
3
Claimed Contributions
26
Contribution Candidate Papers Compared
0
Refutable Paper

Research Landscape Overview

Core task: Pseudo GPS trajectory generation. The field encompasses a diverse set of approaches for creating synthetic movement data that mimics real-world GPS traces. At the highest level, the taxonomy reveals several major branches: Privacy-Preserving Trajectory Synthesis focuses on protecting user identity while maintaining statistical utility; Deep Learning-Based Trajectory Generation leverages neural architectures such as GANs, VAEs, and more recently flow matching and diffusion models to learn complex mobility patterns; Domain-Specific Trajectory Synthesis tailors generation methods to particular contexts like delivery logistics or bicycle routing; Rule-Based and Map-Driven Trajectory Generation relies on explicit constraints and road network structure; Trajectory Analysis and Auxiliary Methods provide supporting techniques for segmentation and feature extraction; while Autonomous Navigation and Control Applications and Route Optimization and Planning address real-time path planning and control problems. Representative works span from early rule-based systems like DPT[9] to modern deep generative models such as Synmob[5] and SynthCAT[6], illustrating the field's evolution toward data-driven, scalable synthesis. Within the deep learning branch, a particularly active line of work explores flow matching and diffusion models, which offer stable training dynamics and high-fidelity sample generation. TrajFlow[0] sits squarely in this emerging cluster, employing flow-based generative modeling to produce realistic trajectories. Its closest neighbor, Geo-lucid Diffusion[11], similarly adopts diffusion techniques but may emphasize different conditioning strategies or geographic priors. Both contrast with earlier GAN-based approaches like Synmob[5], which can suffer from mode collapse, and with privacy-centric methods such as DPT[9] that prioritize differential privacy guarantees over generative flexibility. Meanwhile, domain-specific efforts like Delivery Time Selection[4] and Bicycle Route Synthesis[13] demonstrate how task constraints shape generation, highlighting an ongoing tension between general-purpose generative models and specialized, application-driven synthesis. TrajFlow[0] thus represents a shift toward leveraging modern probabilistic frameworks to balance realism, diversity, and computational efficiency in trajectory generation.

Claimed Contributions

First flow-matching-based generative framework for GPS trajectory modeling

The authors introduce TrajFlow, the first application of flow matching models to GPS trajectory generation. They demonstrate that this paradigm provides improved robustness and stability when generating trajectories across multiple spatial scales compared to diffusion-based approaches.

10 retrieved papers
Unified framework integrating trajectory harmonization, OD-conditioned normalization, and flow-based training

The authors propose a methodological design that combines trajectory harmonization using the RDP algorithm, origin-destination conditioned normalization, and flow matching training. This unified approach simultaneously tackles the challenges of multi-scale generation, transportation-mode diversity, and computational efficiency.

6 retrieved papers
First nationwide multi-scale GPS trajectory generation model

The authors present TrajFlow as the first model capable of generating GPS trajectories at nationwide scale while maintaining performance across urban, metropolitan, and nationwide spatial levels. They validate this using a nationwide mobile phone GPS dataset with millions of trajectories across Japan.

10 retrieved papers

Core Task Comparisons

Comparisons with papers in the same taxonomy category

Contribution Analysis

Detailed comparisons for each claimed contribution

Contribution

First flow-matching-based generative framework for GPS trajectory modeling

The authors introduce TrajFlow, the first application of flow matching models to GPS trajectory generation. They demonstrate that this paradigm provides improved robustness and stability when generating trajectories across multiple spatial scales compared to diffusion-based approaches.

Contribution

Unified framework integrating trajectory harmonization, OD-conditioned normalization, and flow-based training

The authors propose a methodological design that combines trajectory harmonization using the RDP algorithm, origin-destination conditioned normalization, and flow matching training. This unified approach simultaneously tackles the challenges of multi-scale generation, transportation-mode diversity, and computational efficiency.

Contribution

First nationwide multi-scale GPS trajectory generation model

The authors present TrajFlow as the first model capable of generating GPS trajectories at nationwide scale while maintaining performance across urban, metropolitan, and nationwide spatial levels. They validate this using a nationwide mobile phone GPS dataset with millions of trajectories across Japan.

Table of Contents