Survey of Edge Case Detection in Automated Driving

A recent study published on arXiv offers an extensive overview of edge case detection in automated driving, tackling the issue of infrequent and unforeseen scenarios that may jeopardize the dependability of automated vehicles (AVs). Cited as arXiv:2410.08491v3, the research presents a tiered classification of detection and evaluation techniques, organized initially by AV components (including perception and trajectory-related subsystems such as prediction, planning, and control) and subsequently by foundational methodologies and theories. Importantly, the authors introduce "knowledge-driven" strategies that enhance data-driven approaches by utilizing expert insights and domain expertise to uncover scenarios not represented in training datasets. This survey addresses a notable gap in existing literature, as earlier studies lacked a thorough examination of these methods. The paper can be accessed at https://arxiv.org/abs/2410.08491.