Carles Sierra (IIIA-UAB); Dongmo Zhang (WSU); Simeon Simoff (WSU); Dave De Jonge (IIIA-UAB)

In the past decade, the research on autonomous vehicles (AVs) has made revolutionary progress. The advancements in Artificial Intelligence (AI), and especially machine learning, allow self-driving cars to learn how to handle complex road situations based
on data from millions of accumulated driving hours, much more than any human driver could ever reach. Autonomous driving brings us hope for safer, more convenient, more efficient, and more environmentally friendly transportation. However, autonomous vehicles
on roads also introduce new challenges to traffic management. New theories for a better understanding of the new era of transportation and new technologies for smart roadside infrastructures and intelligent traffic control are crucial for the development and deployment of autonomous vehicles as well as human communities.

This thesis aims to take on the challenges to address some of the key issues in traffic control and management, including intersection protocol design, congestion measurement, selfish routing and road infrastructure automation, under the assumption that all vehicles on the road are connected and self-driving.

To design and test traffic control mechanisms for AVs, we introduced a formal model to represent road networks and traffic. Based on this model, we developed a simulation system on top of an existing open-source platform (AIM4) and used it to examine a number of traffic management protocols specifically designed for traffic with fully autonomous vehicles. Simulation outcomes show that traffic management protocols for AVs can be more subtle, sensitive and variable with traffic volumes/flow rate, vehicle safe distance and road configuration. In addition, by analyzing the real-world traffic data and simulation data, we found that measuring congestion with exponential functions has considerable advantages against the traditional BPR function in certain aspects.

The deployment of autonomous vehicles provides traffic management with an opportunity of choosing either centralised control or decentralised control. The price of anarchy (PoA) of autonomous decision-making for routing gives an applicable quantitative criterion for selection between them. We extended the existing research on PoA with the class of exponential functions as cost functions. We found an expression for the tight upper bound of the PoA for selfish routing games with exponential cost functions. Unlike existing studies, this upper bound depends on traffic demands, with which we can get a more accurate estimation of the PoA. Furthermore, by comparing the upper-bounds of PoA between the BPR function and the exponential function, we found that the exponential functions yield a smaller upper bound than the BPR functions in relatively low traffic flows.

To specify traffic management systems with autonomous roadside facilities, we propose a hybrid model of traffic assignment. This model aims to describe traffic management systems in which both vehicles and roadside controllers make autonomous decisions, therefore, are autonomous agents. We formulated a non-linear optimization problem to optimize traffic control from a macroscopic view of the road network. To avoid the complex calculations required for non-linear optimization, we proposed an approximation algorithm to calculate equilibrium routing and traffic control strategies. The simulation results show that this algorithm eventually converges to a steady state. The traffic control scheme in this steady state is an approximately optimal solution.