Towards Safe, Explainable, and Regulated Autonomous Driving

AI alone is insufficient for safe autonomous vehicles: The paper notes that despite empirical successes in deep learning and reinforcement learning, recent traffic accidents demonstrate that current autonomous driving technology "is not fully reliable for safe deployment," indicating that powerful AI techniques require additional safeguards.

Three pillars are essential for responsible autonomous driving: The framework integrates autonomous control, explainable AI (XAI), and regulatory compliance as interconnected requirements—suggesting that safety requires not just technical performance, but also transparency and adherence to regulatory standards.

Explainability is a regulatory necessity, not optional: The paper emphasizes that both stakeholders and transportation regulators require AI-driven software to be "explainable" alongside being safe and compliant, indicating that regulators recognize the need for humans to understand why autonomous vehicles make critical decisions.

Current approaches need structural redesign: Rather than incremental improvements to existing systems, the authors propose an integrated design framework, suggesting that retrofitting explainability and compliance onto existing autonomous driving systems may be insufficient.

XAI methods exist to support safer deployment: The paper describes "relevant XAI approaches" that can help achieve safety and regulatory goals, indicating that practical tools are available to make autonomous driving systems more transparent and trustworthy.