Modular autonomous intersection management simulation for stochastic and priority auction paradigms

Automated intersections, when combined with the proliferation of autonomous vehicles (AVs), allow for more precise and innovative methods to control traffic at these integral choke points in the road system. In this dissertation, I develop a refined, modular framework for autonomous intersection management (AIM) simulation and implement it as a software library with robust documentation and testing to support present and future research in this field. Demonstrating this framework's efficacy, I apply it to study two topic areas in the AIM space: stochastic movement and priority auctions. Stochastic AIM is introduced as an extension of traditional AIM that permits probabilistic reservations of space and time in an intersection. Its use case is motivated by the integration of human-driven vehicles into AIM using augmented reality guidance to behave more accurately to AV movement, while still making some stochastic deviations from AV-identical trajectories. These deviations are quantified using experimental data from human drivers in a driving simulator merged into a stochastic vehicle movement model. Experimental results suggest that, with this paradigm, AIM can decrease delay significantly, even at low AV penetration levels (less than 20%). Finally, I conceptualize intersection priority auctions into the newly developed AIM framework as itself a modular framework that supports the dispatch of multiple vehicles simultaneously from either separate lanes or a single lane without relying on preset signal phases. This auction framework further supports three payment formulas for the winner of the priority auction: first-price, second-price, or a novel externality payment mechanism. Using experiments implemented in the novel AIM simulator, my results demonstrate significant reduction in value-weighted delay using the multiple dispatch configuration and novel payment mechanism compared to other configurations, with the novel formula incentivizing truthful reporting of valuations more than its alternatives.