Diffusion Models for Safety Validation of Autonomous Driving Systems

Juanran Wang *, Marc R. Schlichting *, Harrison Delecki, Mykel J. Kochenderfer · 2025

Read Diffusion Models for Safety Validation of Autonomous Driving Systems by Juanran Wang *, Marc R. Schlichting *, Harrison Delecki, Mykel J. Kochenderfer (2025) 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-2506-08459.txt.

Description

Juanran Wang, Marc Schlichting, Harrison Delecki, and Mykel Kochenderfer train a diffusion model to imagine the ways a self-driving car can fail. Testing on real roads is dangerous and expensive, and failures are rare and varied. Writing in twenty twenty-five, the team conditions the generator on any starting traffic state. It then produces plausible failure scenarios for the vehicle under test. Experiments center on a four-way intersection, a classic conflict zone. The generated samples looked realistic while covering a wide spread of failure types.

No external training data set is required. Training and inference run on modest hardware. The method assumes nothing about the driving system's internals, so it can audit black-box stacks. The authors pitch it as a general safety validation tool for intersections and beyond.

Catalog record

CREATORJuanran Wang *, Marc R. Schlichting *, Harrison Delecki, Mykel J. Kochenderfer
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE29 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-08459

Browse the full library · About the Tower of Babel