Making autonomous vehicle systems human-like: lessons learned from accident experiences in traffic

Abstract

ABSTRACT The COVID-19 pandemic has hastened the adoption of autonomous vehicles (AVs) to minimise human-to-human contact. Yet, prior investigations suggest that AVs are accident-prone when they behave differently from humans. It is necessary to design an autonomous vehicle system (AVS) that can take human behaviour into account. This study capitalises on the wealth of data from traffic accidents caused by humans and discovers association rules to improve AVSs. Findings show that fatal accidents likely co-occur with “right near”, “head on” or “lane side swipe” scenarios. They provide important implications for designing traffic scenarios that are critical for training an AVS.