CRASH: Cognitive Reasoning Agent for Safety Hazards in Autonomous Driving

2026-03-16

AI Summary

This study addresses the challenge of systematically analyzing root causes in autonomous vehicle (AV) accidents, which is hindered by the high heterogeneity of AV system architectures and the lack of standardized assessment frameworks. Leveraging 2,168 real-world AV crash reports from NHTSA (2021–2025), the authors propose a unified representation framework that integrates a cognitive reasoning agent with a large language model to automatically parse both structured and unstructured accident records. The system generates concise summaries, identifies primary contributing factors, and determines whether the AV was a substantial cause of the incident. This approach enables cross-architecture, interpretable, and scalable safety analysis. Experimental results show that 64% of accidents stem from perception or planning failures, with approximately half involving rear-end collisions. Expert validation by five specialists confirms an 86% accuracy rate in fault attribution.

Technology Category

• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning
• Natural Language Processing: Safety and Robustness
• Computer Vision: Vision for Robotics & Autonomous Driving

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Web query analysis, representation and understanding
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

As AVs grow in complexity and diversity, identifying the root causes of operational failures has become increasingly complex. The heterogeneity of system architectures across manufacturers, ranging from end-to-end to modular designs, together with variations in algorithms and integration strategies, limits the standardization of incident investigations and hinders systematic safety analysis. This work examines real-world AV incidents reported in the NHTSA database. We curate a dataset of 2,168 cases reported between 2021 and 2025, representing more than 80 million miles driven. To process this data, we introduce CRASH, Cognitive Reasoning Agent for Safety Hazards, an LLM-based agent that automates reasoning over crash reports by leveraging both standardized fields and unstructured narrative descriptions. CRASH operates on a unified representation of each incident to generate concise summaries, attribute a primary cause, and assess whether the AV materially contributed to the event. Our findings show that (1) CRASH attributes 64% of incidents to perception or planning failures, underscoring the importance of reasoning-based analysis for accurate fault attribution; and (2) approximately 50% of reported incidents involve rear-end collisions, highlighting a persistent and unresolved challenge in autonomous driving deployment. We further validate CRASH with five domain experts, achieving 86% accuracy in attributing AV system failures. Overall, CRASH demonstrates strong potential as a scalable and interpretable tool for automated crash analysis, providing actionable insights to support safety research and the continued development of autonomous driving systems.

Problem

• autonomous driving
• safety hazards
• incident investigation
• fault attribution
• system heterogeneity

Innovation

• LLM-based agent
• automated crash analysis
• cognitive reasoning
• autonomous driving safety
• incident attribution