A Spatial Analysis of Connected Autonomous Vehicle (CAV) Safety

Abstract

Autonomous vehicles (AVs) are poised to be the biggest disruption in transportation technology in decades, and there are already limited commercial AV operations on the road today. Their primary benefit to society is theorized to be increased safety relative to human-driven vehicles (HDVs). Connected autonomous vehicle (CAV) technology has sought to amplify this safety dividend by sharing critical perception information among AVs to address sensor limitations. This poster will provide a spatial analysis of CAV safety outcomes across simulated real-world road environments. CAVs will be simulated using the high-fidelity AV simulation software CARLA and the robust traffic management simulator SUMO. I will simulate realistic traffic scenarios in portions of a southern California city based on historical traffic flows as well as driving behavior models that account for human error (e.g. aggressive driving) and machine sensor error (e.g. communication latency). The analysis will explore spatial patterns in safety outcomes between two variables: the percentage of CAVs on the road, and the state-of-the-art cooperative perception algorithm behind CAV behavior. I will also simulate AVs not using cooperative perception algorithms to compare AV safety when perception remains siloed. Safety outcomes will be measured via surrogate safety measures (SSMs) that record the location and severity of collisions or unsafe near-collisions. This effort will critically evaluate the safety advantage of CAV technology, and explore spatial patterns in remaining safety incidents to see where further improvements can be made.