A Data-Driven Autonomous Driving System for Overtaking Bicyclists

Abstract

This research aims to develop data-driven models for suggesting the initiation of an automated car-to-bicycle overtaking process that will be assessed subjectively by human drivers and bicyclists in a driving simulator environment. A naturalistic driving dataset with 102 vehicles involved served as the data source for model development. The models were implemented to a CarSim software as the driving simulator platform for an experiment. Thirty-two participants were recruited to evaluate the models from driver and bicyclist’s perspectives on the aspects of satisfaction and perceived risk of collision. It was found that both drivers and bicyclists felt less satisfied and perceived higher risk if the overtaking was engaged with a faster speed and the presence of incoming traffic. However, the effect to bicyclists could be mitigated with the application of a dedicated bicycle lane. Bicyclists also sought more lateral room to the vehicle when being overtaking, although drivers were satisfied with the current settings without perceiving any significant risk. Therefore, the developed models should be adjusted in the future by considering the perceptions by bicyclists and other road users. Stakeholders, such as automated feature developers and policymakers, should referto the models carefully with paying attention to the inconsistency between driver’s and bicyclist’s perspectives.