Car following dynamics in mixed traffic flow of autonomous and human-driven vehicles: Complex networks approach

Abstract

Autonomous driving technologies have demonstrated exceptional performance in improving traffic operational efficiency and safety, contributing to the growing market penetration rate of autonomous vehicles (AVs). This study focuses on analyzing the interaction between AVs and human-driven vehicles (HVs) in mixed traffic flow, with an emphasis on the behavioral differences among various car-following (CF) vehicle pair types. While previous research has primarily relied on simulation and statistical methods to quantify the interaction between AVs and HVs, these approaches might overlook real-world driving nuances and fail to capture the dynamic changes in driving behavior. To address the limitations, we utilize a mixed traffic flow dataset (i.e., Lyft Level-5 Open Dataset), and apply a coarse-grained phase-space algorithm to model the dynamic changes in CF behavior. The interactions of different vehicle pairs are represented as directed, weighted complex networks. By analyzing network metrics, extracting core subgraphs, and calculating network similarities, the result indicates that the type of car following vehicle pair significantly influences following behavior. Moreover, changes in the leading or following vehicles within a platoon can lead to shifts in following behavior, and the introduction of AVs contributes positively to enhancing both the safety and efficiency of traffic flow. These network-based findings enrich the understanding of interactions between different vehicle types in mixed traffic flow and provide a solid foundation for designing mixed traffic flow control algorithms that account for vehicle type heterogeneity.