Conditional Flow Matching for Continuous Anomaly Detection in Autonomous Driving on a Manifold-Aware Spectral Space
Antonio Guillen-Perez · 2026
Read Conditional Flow Matching for Continuous Anomaly Detection in Autonomous Driving on a Manifold-Aware Spectral Space by Antonio Guillen-Perez (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-17586.txt.
Description
Antonio Guillen-Perez tackles a bottleneck in self-driving safety. Rare, high-risk edge cases slip past rule-based checks, so Level 4 cars lack a scalable way to prove they are safe. His framework, Deep-Flow, learns the density of expert human driving with no labels, then flags acts that fall outside normal road conduct.
The method uses optimal transport conditional flow matching, squeezed through a low-rank spectral manifold. That keeps motion smooth and likelihoods stable. A transformer with lane-aware goal conditioning handles tricky junctions. A weighting scheme favors energetic moves in training. On Waymo data, it surfaced lane and junction breaches that standard filters missed, pointing to data-driven safety gates.
Catalog record
| CREATOR | Antonio Guillen-Perez |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 59 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-17586