CCFM: Collision-Constrained Flow Matching for Safety-Critical Scenario Generation

2026-07-05

AI Summary

Existing methods struggle to controllably generate safety-critical driving scenarios that satisfy specific collision types and severity constraints. This work proposes the first flow-matching framework incorporating hard physical constraints to precisely govern four collision categories: rear-end, side-impact, cut-in, and head-on collisions. The approach integrates composite scoring heuristics for adversarial agent selection, structured collision constraints based on contact points, heading angles, and severity levels, and a constrained flow-matching sampler employing Gauss–Newton manifold projection. Evaluated on nuScenes and nuPlan, the method achieves collision generation rates of 46.4% and 83.1%, respectively—significantly outperforming baseline approaches—while preserving realistic driving behaviors. By enforcing hard constraints rather than relying on soft guidance, this framework overcomes the probabilistic limitations inherent in prior methods.

Abstract

Evaluation of autonomous vehicle (AV) planners in safety-critical closed-loop simulation is essential for real-world deployment. However, generating controllable safety-critical scenarios remains challenging. Existing approaches use soft guidance that provides only probabilistic preferences and cannot guarantee the satisfaction of geometric and severity constraints associated with specific collision types. We introduce Collision-Constrained Flow Matching (CCFM), a novel framework that guarantees precise collision control through hard physical constraints. CCFM consists of three key components: (i) a heuristic collision selector that optimally identifies an adversarial agent and collision type via composite scoring; (ii) structured hard constraints that explicitly define four collision types (rear-end, side, cut-in, head-on) through contact point, heading, and severity requirements; and (iii) a collision-constrained flow matching sampler that enforces the constraints via Gauss-Newton manifold projection. CCFM achieves collision rate up to 46.4% on nuScenes and 83.1% on nuPlan, significantly outperforming baselines while preserving realistic driving behavior. By enabling controllable collision characteristics in safety-critical scenario generation, CCFM provides a reliable foundation for AV safety evaluation and sim-to-real crash data generation. The code and implementation details are available at https://github.com/KELISBU/CCFM.

Problem

• safety-critical scenario generation
• collision constraints
• autonomous vehicle evaluation
• controllable collision
• closed-loop simulation

Innovation

• Collision-Constrained Flow Matching
• hard constraints
• safety-critical scenario generation
• Gauss-Newton manifold projection
• autonomous vehicle evaluation