Coordinated Transportation Systems: From Grid-Savvy Routing For Electric Trucks To Stochastic Control Of Demand-Responsive Vehicles

 The coordination of vehicles for the movement of goods and people is crucial for achieving economically viable, sustainable, and equitable transportation systems. This thesis explores two settings of optimally planning and controlling fleets of vehicles: one bridges the gap in electrifying delivery services, while the other enables demand-responsive and equitable public transportation. Focusing first on the economic viability of grid-savvy electric delivery fleets, I develop planning algorithms that optimize fleet design, route planning, and managed charging. This addresses the technological constraints limiting the adoption of electrified medium- and heavy-duty vehicles (MHDVs) that make them fundamentally different to operate than their conventional fossil-fueled counterparts. I extend the electric vehicle routing problem to jointly optimize routing and vehicle-to-grid interactions. Through regional case studies, I demonstrate that EVs can become cost-competitive with diesel trucks when coordinating delivery routes and charging under time-varying commercial electricity rates. Furthermore, EVs with vehicle-to-grid (V2G) capabilities can reduce peak demand charges and exploit price differences between utilities, resulting in net negative costs in some cases. The second setting in this dissertation demonstrates how optimal coordination of a demand-responsive vehicle with shared transportation provides service quality and equity improvements in passenger transportation. First, I propose a sequential decision-making problem in which a demand-responsive vehicle selects actions to coordinate the service of pooled ride requests on routes with shared transit to maximize ridership, reduce travel times, and increase service quality while considering costs and equity preferences specified by a centralized transit agency. Several heuristic methods, exact dynamic programming, and reinforcement learning methods are investigated to optimally control the demand-responsive vehicle. Through a New York City case study, I demonstrate the feasibility of each method for operating a demand-responsive vehicle that augments shared transit to maximize equity and service quality metrics. The results validate the viability of reinforcement learning approaches compared to heuristic policies at realistic scales. The contributions of this thesis provide insights into the operational challenges of coordinating fleets of vehicles across different transportation domains and offer algorithmic solutions to enable more economically viable and equitable mobility systems.