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

2026-03-04

AI Summary

Existing methods for generating safety-critical scenarios in autonomous driving struggle to simultaneously ensure adversarial effectiveness, physical feasibility, and behavioral realism. This work proposes SaFeR, which formulates scenario generation as a discrete next-token prediction problem. Built upon a Transformer architecture, SaFeR introduces a differential attention mechanism to suppress noise and integrates a Largest Feasible Region (LFR) constraint with a feasibility-guided token resampling strategy to efficiently induce adversarial behaviors within high-probability realistic regions. The LFR is approximated via offline reinforcement learning. Closed-loop evaluations on the Waymo Open Motion Dataset and nuPlan demonstrate that SaFeR significantly outperforms existing approaches in terms of solvability rate, kinematic realism, and adversarial efficacy.

Technology Category

• Natural Language Processing: Safety and Robustness
• Computer Vision: Adversarial Attacks & Robustness
• Multiagent Systems: Adversarial Agents

Application Category

• Search and Retrieval-Augmented AI: Agentic search
• Security and Privacy: Large-scale security measurements
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Safety-critical scenario generation is crucial for evaluating autonomous driving systems. However, existing approaches often struggle to balance three conflicting objectives: adversarial criticality, physical feasibility, and behavioral realism. To bridge this gap, we propose SaFeR: safety-critical scenario generation for autonomous driving test via feasibility-constrained token resampling. We first formulate traffic generation as a discrete next token prediction problem, employing a Transformer-based model as a realism prior to capture naturalistic driving distributions. To capture complex interactions while effectively mitigating attention noise, we propose a novel differential attention mechanism within the realism prior. Building on this prior, SaFeR implements a novel resampling strategy that induces adversarial behaviors within a high-probability trust region to maintain naturalism, while enforcing a feasibility constraint derived from the Largest Feasible Region (LFR). By approximating the LFR via offline reinforcement learning, SaFeR effectively prevents the generation of theoretically inevitable collisions. Closed-loop experiments on the Waymo Open Motion Dataset and nuPlan demonstrate that SaFeR significantly outperforms state-of-the-art baselines, achieving a higher solution rate and superior kinematic realism while maintaining strong adversarial effectiveness.

Problem

• safety-critical scenario generation
• autonomous driving test
• physical feasibility
• behavioral realism
• adversarial criticality

Innovation

• feasibility-constrained resampling
• safety-critical scenario generation
• realism prior
• differential attention
• largest feasible region