A Taxonomy and Systematic Review of Driving Automation Failure Scenarios

Before human drivers are fully removed from the vehicle control loop, transitioning control between automation and human drivers in safety-critical scenarios remains a major challenge. Although existing research has proposed several taxonomies to categorize safety-critical scenarios from technical and human-factors perspectives, gaps remain. Thus, we propose a framework compatible with previous taxonomies from multiple perspectives. The taxonomy includes three dimensions: the source of harm, the urgency level, and the trigger conditions. We then conducted a systematic literature review of safety-critical scenarios in existing empirical studies (N = 292) using our proposed taxonomy. We identified several major research gaps, including the highly unbalanced scenario types, limited understanding of information needs, and a lack of appropriate strategies to improve takeover performance in specific scenarios. These findings underscore the need for a unified framework to categorize and analyze safety-critical scenarios, supporting researchers, practitioners, and policymakers in testing and improving the safety of automated vehicles.