Explainable Artificial Intelligence for Autonomous Driving: A Comprehensive Overview and Field Guide for Future Research Directions

TL;DR

This paper surveys XAI for autonomous driving and argues that safety and regulatory acceptance depend on transparent AI decisions. It frames the driving stack as an end-to-end controller coupled with a safety-regulatory component and an explanations module, then reviews methods across visual explanations, reinforcement-learning and imitation-learning explanations, decision-tree and logic-based approaches, user studies, and large-language-model/vision-language-model explanations. It introduces a unifying XAI framework that integrates end-to-end control, safety verification, and explanations, and discusses AV2.0 with Embodied AI that unifies vision, language, and action while addressing safety challenges and hallucination risks. The paper highlights temporal granularity of explanations, human-machine interfaces design, and stakeholder-specific explanations as key directions to realize trustworthy, regulatory-compliant AVs.

Abstract

Autonomous driving has achieved significant milestones in research and development over the last two decades. There is increasing interest in the field as the deployment of autonomous vehicles (AVs) promises safer and more ecologically friendly transportation systems. With the rapid progress in computationally powerful artificial intelligence (AI) techniques, AVs can sense their environment with high precision, make safe real-time decisions, and operate reliably without human intervention. However, intelligent decision-making in such vehicles is not generally understandable by humans in the current state of the art, and such deficiency hinders this technology from being socially acceptable. Hence, aside from making safe real-time decisions, AVs must also explain their AI-guided decision-making process in order to be regulatory compliant across many jurisdictions. Our study sheds comprehensive light on the development of explainable artificial intelligence (XAI) approaches for AVs. In particular, we make the following contributions. First, we provide a thorough overview of the state-of-the-art and emerging approaches for XAI-based autonomous driving. We then propose a conceptual framework that considers the essential elements for explainable end-to-end autonomous driving. Finally, we present XAI-based prospective directions and emerging paradigms for future directions that hold promise for enhancing transparency, trustworthiness, and societal acceptance of AVs.

Figures (6)

• Figure 1: A canonical example of explainable AI in autonomous driving: An autonomous vehicle provides a live natural language explanation of its real-time decision to bystanders. The image has been adapted and modified from the original source: daimler2017.
• Figure 3: Taxonomy of the stakeholders in autonomous driving.
• Figure 4: An example of a counterfactual explanation generated by STEEX. Graphics credit: jacob2022steex.
• Figure 5: Human advice to the car for relevant action. Source: kim2020advisable.
• Figure 6: RL-based interpretable end-to-end autonomous driving via a bird-eye mask. Credit: chen2021interpretable