How Effective are Identification Technologies in Autonomous Driving Vehicles?

Abstract

Autonomous driving necessarily involves timely awareness of surrounding environmental conditions to facilitate safe navigation. Vision is therefore of paramount importance in these vehicles. Cameras, LiDAR, RADAR, and GNSS provide a reasonable amount of necessary environmental input in a majority of current autonomous driving initiatives. Several published studies vouch for the advantages of autonomous vehicles over their human-driven counterparts, in principle. However, extant literature does not provide clear guidance on the extent of dominance, if any, of autonomous vehicles in terms of accident avoidance. We consider ‘vision’ inputs in autonomous vehicles and compare their performance to that of human-driven vehicles based on recent accident data and show that current state-of-the-art of vision technology in automated vehicles are grossly insufficient for truly autonomous vehicles. Specifically, our results illustrate the extent of deficit that must be addressed in state-of-the-art machine learning algorithms and vision sensors that are used in autonomous driving vehicles.