Enabling the next generation of transportation systems by accounting for heterogeneity in traffic flow: Modeling and control of mixed autonomy traffic

Abstract

Traffic engineering is a field characterized by heterogeneity, reflecting the diverse behaviors of individual agents using the infrastructure. While traffic heterogeneity has been discussed for several decades, primarily focusing on vehicle types and human driver behavior, recent years have seen an expansion of the conversation to include emerging technologies like automated and electric vehicles. This heterogeneity impacts driving behavior such that even a single agent's actions can substantially influence local traffic flow. Motivated by such findings, this dissertation focuses on understanding and investigating the heterogeneity of driving behavior in traffic flow, particularly the intrinsic differences between autonomous driving and human driving and their impacts on system-level dynamics. The goal is to leverage these differences to improve overall performance, sustainability, and resilience. From the standpoint of traffic flow, altering the behavior of even a small number of individual agents can have significant implications for the emergent properties of the entire flow, such as passenger travel time and vehicle energy consumption. If properly controlled, certain desirable properties of traffic flow, such as stability and increased highway throughput, can be achieved. The spectrum of vehicle automation spans from SAE Level 1 – comprising partially automated vehicles equipped with driver-assist functions like adaptive cruise control (ACC) – to SAE Level 5, which refers to fully automated vehicles (AVs) that operate without any human intervention. While numerous benefits have been demonstrated for fully automated traffic flow, it remains unclear how partially automated vehicles, which are already commercially available, will influence mixed autonomy traffic characteristics in the near future. Therefore, this dissertation aims to develop methodological tools for the mathematical modeling, simulation, and control of mixed autonomy traffic involving fully automated, partially automated, and human-driven vehicles. This dissertation fits into three distinct thrusts: i) developing physically interpretable car-following models to accurately describe the dynamics of ACC vehicles and human-driven vehicles (Chapters 2, 3, and 4); ii) investigating how commercially available ACC vehicles will impact mixed autonomy traffic with human-driven vehicles at different market penetration rates (Chapters 5 and 6); iii) designing and controlling next-generation intelligent infrastructure and vehicles to better adapt to mixed autonomy traffic (Chapters 7, 8, and 9).