Autonomous vehicle lane-changing dynamics and impact on the immediate follower

Abstract

• Developed random parameters linear regression and extreme value models for quantifying lane-changing. • Autonomous vehicle lane-change causes greater speed reduction than human-driven vehicles. • Lane-change impact is significantly related to lane-changing decision-making and execution. • Follower’s crash risk significantly higher when lane-changer is an autonomous vehicle. Understanding and modelling lane-changing behaviour are critical aspects of microscopic traffic flow modelling, safety analyses, and microsimulation due to their significant impact on traffic flow characteristics and safety. Among the three aspects of lane-changing behaviour—decision-making, execution, and impact—the lane-changing impact has been comparatively underexplored in the literature, which is disproportionate to its importance. A lack of proper understanding of lane-changing impact may lead to inaccurate planning and interpretation of mixed traffic stream comprising both autonomous and human-driven vehicles. Motivated by this research gap, the current study investigates the lane-changing impact of autonomous vehicles on the immediate follower using the publicly available Waymo Open Dataset. Human-driven vehicle lane-changing data are also extracted from the same database and used for comparison. Lane-changing impact on traffic flow efficiency and safety is examined through the speed reduction of the follower in the target lane and deceleration rate to avoid a collision for the same follower, respectively. A correlated random parameters linear regression model is employed to assess the speed reduction of the follower as a function of lane-change duration, lag gap, lane-changer speed, and a dummy variable indicating whether the lane-changer is an autonomous vehicle or a human-driven vehicle. The results reveal that lane changes executed by autonomous vehicles may cause greater or lesser speed reductions for the follower compared to those executed by human-driven vehicles, which could be attributed to the heterogeneous behaviour of followers perceiving and responding differently to autonomous vehicle lane-changes compared to human-driven ones. Further the block maxima and peak over threshold models are developed to estimate crash risk for the follower in the target lane using a deceleration rate to avoid a collision conflict measure. The results suggest that the risk of a collision increases substantially when the lane-changer is an autonomous vehicle. This elevated risk may be associated with drivers’ lack of trust in autonomous vehicles and traffic dynamics, reflecting self-inflicting hard deceleration to avoid potential collisions. Overall, this study highlights the heterogeneous impacts of lane-changing by autonomous vehicles on the immediate follower, emphasising the need for tailored models that accurately capture the dynamics of surrounding traffic behaviour. The findings will be helpful to road safety engineers and policymakers in planning mixed traffic with the safe integration of autonomous vehicles.