KG-ASG: Collision-Knowledge-Guided Closed-Loop Adversarial Scenario Generation With Primary-Support Attribution

2026-05-17

AI Summary

This work addresses key challenges in generating high-risk scenarios for autonomous driving safety validation—namely, ambiguous collision semantics, uncontrolled multi-vehicle collisions, and poor executability of adversarial samples—by proposing a collision knowledge-guided leader-follower attribution mechanism. The approach leverages a structured collision knowledge base and a lightweight Collision Expert model to infer target collision patterns and assign leader-follower roles among vehicles, modeling multi-agent adversarial interactions as a process where the leader initiates conflict while followers shape the risk environment without directly colliding. Hard constraints grounded in traffic rules, physical feasibility, and interactive safety filter invalid samples, and a planning-control closed-loop feedback loop enables failure diagnosis and sample refinement. Evaluated on MetaDrive-reconstructed WOMD scenarios, the method significantly improves effective leader attack rates, reduces multi-collision incidence, and enhances closed-loop recovery performance across diverse controllers.

Abstract

Safety validation of autonomous driving systems requires high-risk scenario coverage, clear collision semantics, executable trajectories, and attributable multi-vehicle interactions. Existing safety-critical scenario generation methods often rely on low-level trajectory perturbations, collision-proxy optimization, or single-adversary search, which may produce adversarial samples with ambiguous collision causes or uncontrolled multi-vehicle collisions. This paper proposes KG-ASG, a collision-knowledge-guided closed-loop adversarial scenario generation framework with primary-support attribution. KG-ASG constructs a structured collision knowledge base and trains a lightweight Collision Expert to infer the target collision mode, the unique primary adversary, support vehicles, and their interaction roles. Guided by this semantic prior, multi-vehicle adversarial generation is formulated as a primary-support process, where the primary adversary induces the main conflict and support vehicles shape the surrounding risk structure without becoming additional colliders. Rule, physical, interaction-safety, and single-collider constraints are imposed as hard gates to filter non-executable samples. To handle reactive ego behaviors, planner-controller feedback is further used for failure diagnosis, candidate re-ranking, and terminal refinement. Experiments on WOMD scenarios reconstructed in MetaDrive show that KG-ASG achieves strong adversarial effectiveness while improving Valid Primary Attack, reducing multi-collision, and obtaining closed-loop recovery gains under IDM, Cruise, and Expert controllers. These results demonstrate that collision-knowledge guidance and primary-support single-collider reasoning improve adversarial effectiveness, interpretability, and executability for autonomous driving safety validation.

Problem

• adversarial scenario generation
• collision semantics
• multi-vehicle interaction
• autonomous driving safety validation
• attributable collision

Innovation

• collision-knowledge-guided
• primary-support attribution
• adversarial scenario generation
• closed-loop validation
• single-collider constraint