How corner is a corner case? Percentile control for highway scenario generation
A diffusion method generates highway scenarios at a requested risk percentile, hitting 98.75% of targets.
The paper treats the adversity of a generated autonomous-driving scenario as its percentile in the conditional distribution of future risk given observed history. A reference risk model maps requested percentiles to physical targets, and a percentile-conditioned joint diffusion model with sampling-time risk guidance generates multi-agent futures. Using minimum post-encroachment time on highD, the method realizes 1,422 of 1,440 requests within a 0.05 percentile tolerance (98.75%), with mean percentile error 0.00673 and PET-target error 0.00991 seconds.
- Scenario adversity is defined as a percentile of history-conditioned future risk.
- A percentile-conditioned joint diffusion model generates multi-agent highway futures.
- Risk is measured by minimum post-encroachment time on the highD dataset.
- It met 1,422 of 1,440 requests within a 0.05 percentile tolerance.
- Tags
- autonomous-vehiclesscenario-generationdiffusioncorner-caseshighway-simulationhighdsafety-testing
- Products
- highD
From the original article
1 min read at huggingface.co
Generating corner-case scenarios with appropriate adversity in a simulation environment is critical for testing an autonomous vehicle (AV) software stack's safety performance before deployment.
By Jiaxi Liu, Hang Zhou, Hangyu Li · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.