![[2602.17586] Conditional Flow Matching for Continuous Anomaly Detection in Autonomous Driving on a Manifold-Aware Spectral Space](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.17586] Conditional Flow Matching for Continuous Anomaly Detection in Autonomous Driving on a Manifold-Aware Spectral Space
Summary
This paper presents Deep-Flow, an innovative framework for anomaly detection in autonomous driving, utilizing Optimal Transport Conditional Flow Matching to improve safety validation for Level 4 vehicles.
Why It Matters
As autonomous vehicles become more prevalent, ensuring their safety in rare and complex scenarios is crucial. This research addresses the limitations of traditional methods by introducing a data-driven approach that enhances the detection of high-risk situations, thereby contributing to safer autonomous driving technologies.
Key Takeaways
- Deep-Flow uses Optimal Transport Conditional Flow Matching for anomaly detection.
- The framework effectively identifies high-risk driving behaviors overlooked by traditional methods.
- It employs a PCA bottleneck to ensure kinematic smoothness in generative processes.
- The model achieves a notable AUC-ROC score of 0.766 on the Waymo Open Motion Dataset.
- Deep-Flow distinguishes between kinematic danger and semantic non-compliance, enhancing safety validation.
Computer Science > Robotics arXiv:2602.17586 (cs) [Submitted on 19 Feb 2026] Title:Conditional Flow Matching for Continuous Anomaly Detection in Autonomous Driving on a Manifold-Aware Spectral Space Authors:Antonio Guillen-Perez View a PDF of the paper titled Conditional Flow Matching for Continuous Anomaly Detection in Autonomous Driving on a Manifold-Aware Spectral Space, by Antonio Guillen-Perez View PDF HTML (experimental) Abstract:Safety validation for Level 4 autonomous vehicles (AVs) is currently bottlenecked by the inability to scale the detection of rare, high-risk long-tail scenarios using traditional rule-based heuristics. We present Deep-Flow, an unsupervised framework for safety-critical anomaly detection that utilizes Optimal Transport Conditional Flow Matching (OT-CFM) to characterize the continuous probability density of expert human driving behavior. Unlike standard generative approaches that operate in unstable, high-dimensional coordinate spaces, Deep-Flow constrains the generative process to a low-rank spectral manifold via a Principal Component Analysis (PCA) bottleneck. This ensures kinematic smoothness by design and enables the computation of the exact Jacobian trace for numerically stable, deterministic log-likelihood estimation. To resolve multi-modal ambiguity at complex junctions, we utilize an Early Fusion Transformer encoder with lane-aware goal conditioning, featuring a direct skip-connection to the flow head to maintain intent-integrity throu...