SaFeR: Safety-Critical Scenario Generation for Autonomous Driving Test via Feasibility-Constrained Token Resampling

Jinlong Cui, Fenghua Liang, Guo Yang, Chengcheng Tang, Jianxun Cui · 2026

Read SaFeR: Safety-Critical Scenario Generation for Autonomous Driving Test via Feasibility-Constrained Token Resampling by Jinlong Cui, Fenghua Liang, Guo Yang, Chengcheng Tang, Jianxun Cui (2026) 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-2603-04071.txt.

Description

Cui and others propose SaFeR, feasibility-constrained token resampling for safety-critical driving tests. Existing methods struggle to balance hostile criticality, real feasibility, and behavioral realism. SaFeR treats traffic output as discrete next-token output with a Transformer realism prior and differential focus. Resampling then creates hostile behaviors inside a high-probability trust region under a Largest Feasible Region constraint, approximated offline by reinforcement learning to prevent inevitable collisions. Closed-loop tests on Waymo and nuPlan show SaFeR beating rivals in answer rate and realism with strong hostile effectiveness.

Catalog record

CREATORJinlong Cui, Fenghua Liang, Guo Yang, Chengcheng Tang, Jianxun Cui
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE34502
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.04071

Browse the full library · About the Tower of Babel