UC Berkeley

The issue of vehicular traffic congestion has become a critical challenge in metropolitan areas around the world. Effectively addressing this problem requires a deep understanding of traffic congestion from a behavioral perspective, as human drivers tend to prioritize their own interests, which leads to selfish behaviors that negatively impact the entire traffic system’s efficiency. In contrast, autonomous and connected vehicles, under rapid development, are capable of better coordinating their motion with other neighboring autonomous vehicles and with roadside infrastructure resulting in potential significant capacity and mobility improvements in the overall transportation networks. However, as a consequence, transportation systems are facing not only unprecedented opportunities but also challenges in the transition to future intelligent transportation systems involving autonomous vehicles.

This dissertation explores the potential improvements that autonomous road vehicles may bring in diverse transportation scenarios and their overall impact on the broader transportation landscape. The study focuses on two approaches for the application of autonomous vehicles: first, as altruistic decision-makers, and second, by maintaining a shortened headway.These approaches are analyzed via four scenarios that are typical in road vehicle transportation networks: diverges with a bifurcating lane in the middle, highway on-ramps, vehicles’ routing on networks, and highway toll lanes. The study emphasizes three challenges that are crucial to successfully integrate autonomous vehicles into existing transportation systems that are predominantly transited by human-driven vehicles: first, accurately yet concisely modeling human behavior; second, modeling multi-agent systems that incorporate the key features of autonomous vehicles; and third, developing suitable traffic management and optimization strategies for societal benefits. This dissertation aims to shed light on the complexities of the current transportation revolution and provide valuable insights into the path forward: autonomous vehicles have various potentials to serve for enhanced societal benefits while selfish drivers may exploit the benefits brought by autonomous vehicles, and therefore, effective management and optimization methods are necessary to boost the performance of transportation networks to pave the way for a safer, more efficient, and more sustainable future transportation system.

Specifically, in this dissertation, a unified game-theoretic framework is first presented to model and examine the selfish lane choice behavior of human-driven vehicles at various traffic merges and diverges, which exhibits promising predictive power with minimal parameter calibration requirements. A systematic method is then proposed to induce altruistic decision-making behavior of autonomous vehicles locally, which configures the costs perceived by autonomous vehicles with a socially aware component. Moreover, a comprehensive theoretical analysis is conducted from both static and dynamic perspectives on the routing of mixed autonomy, where autonomous vehicles are configured with a controllable shorter longitudinal headway compared to human-driven vehicles. This analysis examines the impact and stability of the resulting routing system, providing valuable insights into the potential benefits induced by autonomous vehicles with shortened headway. Furthermore, the coexistence of mixed autonomy and high-occupancy vehicles in a toll lane scenario is investigated and a unified toll lane framework that integrates and compares autonomous vehicles and high-occupancy vehicles is proposed. The effectiveness of this framework is demonstrated across various application situations, including toll design, policy formulation and regulationof autonomy.