Research papers · Posted 2026.03.04 · Preprint
SaFeR: Safety-Critical Scenario Generation for Autonomous Driving Test via Feasibility-Constrained Token Resampling
SaFeR: Safety-Critical Scenario Generation for Autonomous Driving Test via Feasibility-Constrained Token Resampling
Overview
SaFeR resamples traffic tokens within a learned realism prior while enforcing feasibility constraints estimated through offline reinforcement learning. Closed-loop evaluations examine adversarial effectiveness alongside physical solvability and naturalistic behavior.
Version: v1