Human-Centric explanations for users in automated Vehicles: A systematic review

Abstract

Background

The decision-making processes of automated vehicles (AVs) can confuse users and reduce trust, highlighting the need for clear and human-centric explanations. Such explanations can help users understand AV actions, facilitate smooth control transitions and enhance transparency, acceptance, and trust. Critically, such explanations could improve situational awareness and support timely, appropriate human responses, thereby reducing the risk of misuse, unexpected automated decisions, and delayed reactions in safety–critical scenarios. However, current literature offers limited insight into how different types of explanations impact drivers in diverse scenarios and the methods for evaluating their quality. This paper systematically reviews what, when and how to provide human-centric explanations in AV contexts.

Methods

The systematic review followed PRISMA guidelines, and covered five databases—Scopus, Web of Science, IEEE Xplore, TRID, and Semantic Scholar—from 2000 to April 2024. Out of 266 identified articles, 59 met the inclusion criteria.

Results

Providing a detailed content explanation following AV’s driving actions in real time does not always increase user trust and acceptance. Explanations that clarify the reasoning behind actions are more effective than those merely describing actions. Providing explanations before action is recommended, though the optimal timing remains uncertain. Multimodal explanations (visual and audio) are most effective when each mode conveys unique information; otherwise, visual-only explanations are preferred. The narrative perspective (first-person vs. third-person) also impacts user trust differently across scenarios.

Conclusions

The review underscores the importance of tailoring human-centric explanations to specific driving contexts. Future research should address explanation length, timing, and modality coordination and focus on real-world studies to enhance generalisability. These insights are vital for advancing the research of human-centric explanations in AV systems and fostering safer, more trustworthy human-vehicle interactions, ultimately reducing the risk of inappropriate reactions, delayed responses, or user error in traffic settings.