Percentile-controlled diffusion generates calibrated corner-case scenarios for AV safety testing
Jiaxi Liu · hf · 2026-10-09
Existing AV scenario generators can enforce behavior, adversity, or feasibility but offer little control over how extreme a generated corner case is relative to plausible futures in the same traffic context.
This work defines a scenario's adversity as its percentile in the conditional distribution of future risk given the observed history. A reference risk distribution maps each requested percentile to a physical risk target; a percentile-conditioned joint diffusion model with sampling-time risk guidance then generates multi-agent futures, with a reference-based criterion for evaluating percentile realization.
Using PET as risk surrogate on highD, the method realizes 1,422 of 1,440 requests within 0.05 percentile tolerance (98.75%), mean percentile error 0.00673 and PET-target error 0.00991 s — connecting context-relative risk specification, physical realization, and evaluation on one scale.
More from Embodied
- Text2Sim turns text prompts into editable physics-grounded videos and sim setups — erwincoumans · 2026-10-09
- Singapore's first giant robot fight flopped because teleop policies weren't built for combat, researcher argues — carlosdponx · 2026-10-09
- PredActor unifies humanoid motion generation and control in one diffusion policy at 50Hz on Unitree G1 — carlosdponx · 2026-10-09
- Meta Glasses US Open ad goes viral: your phone might just stay in your pocket — armand_ruiz · 2026-10-09
- Intel's John Healy: most enterprise robotics value sits in the installed base — ryanshrout · 2026-10-09
- IROS 2026 Best Paper: Feeding 3D Features Directly into VLM Lifts Navigation Success to 74.2% — qinzytech · 2026-10-09