A simulator-trained neural semantic field plus a hierarchical risk tree estimates per-agent collision risk and time-to-collision from monocular video, with foundation-model features used to close the sim-to-real gap without retraining.
UMAD: Unsupervised Mask-Level Anomaly Detection for Autonomous Driving
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
abstract
Dealing with atypical traffic scenarios remains a challenging task in autonomous driving. However, most anomaly detection approaches cannot be trained on raw sensor data but require exposure to outlier data and powerful semantic segmentation models trained in a supervised fashion. This limits the representation of normality to labeled data, which does not scale well. In this work, we revisit unsupervised anomaly detection and present UMAD, leveraging generative world models and unsupervised image segmentation. Our method outperforms state-of-the-art unsupervised anomaly detection.
fields
cs.AI 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
NSF-HRPT: Neural Semantic Field meets Hierarchical Risk Perception Tree for Safety-Critical Scenario Assessment
A simulator-trained neural semantic field plus a hierarchical risk tree estimates per-agent collision risk and time-to-collision from monocular video, with foundation-model features used to close the sim-to-real gap without retraining.