Explainable AI for Safe and Trustworthy Autonomous Driving: A Systematic Review

Explainability is essential for autonomous vehicle safety: The paper identifies XAI (explainable AI) as "fundamental" to meeting safety requirements in autonomous driving, addressing a critical gap where "inscrutable AI systems exacerbate the existing challenge of safety assurance" in self-driving cars.

Five specific XAI approaches can improve trustworthiness: The research identifies five key contributions of XAI for safe autonomous driving: interpretable design, interpretable surrogate models, interpretable monitoring, auxiliary explanations, and interpretable validation—each serving distinct purposes in making AI decisions transparent and verifiable.

Explanations must address data, models, and decision-making authority: The paper emphasizes that trustworthy autonomous driving requires explainability across three critical areas: the data used to train AI systems, the AI models themselves, and the agency (who/what makes decisions), suggesting that transparency must be comprehensive rather than focused on any single component.

A modular framework can balance transparency with safety: The proposed SafeX framework demonstrates that delivering explanations to users doesn't require compromising safety—the system is designed to provide transparency while "simultaneously ensuring the safety of AI models," suggesting these goals can be complementary rather than conflicting.

Current AI systems lack sufficient transparency for public trust: By conducting the "first comprehensive systematic literature review" on this topic, the paper implies that explainability in autonomous driving remains an underdeveloped area, highlighting an urgent need for better practices before widespread deployment.