Predicting Risks Before They Happen: Vision-Language Reasoning for Autonomous Driving
Predicting Risks Before They Happen: Vision-Language Reasoning for Autonomous Driving
This paper introduces PotentialRiskQA, an ego-view driving dataset designed to link observable precursors to potential risks and mitigation actions. The authors also present PR-Reasoner, a vision-language model trained on this dataset that generates precursor-centered reasoning chains. Experiments demonstrate improved scene understanding, risk cognition, and action quality with fewer unsupported risk claims compared to existing methods.
The authors introduce PotentialRiskQA, a new ego-view driving dataset with structured annotations linking precursors to risks and mitigation actions, and PR-Reasoner, a vision-language model trained on this data that improves risk cognition and planning performance on NAVSIM.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
Source：arXiv Robotics — research abstracts · arxiv.org