Date Thesis Awarded
Honors Thesis -- Access Restricted On-Campus Only
Bachelors of Science (BS)
Anke van Zuylen
Through simulation, we demonstrate that incorporation of self-driving vehicles into ride-hailing fleets can greatly improve urban mobility. After modeling existing driver-rider matching algorithms including Uber’s Batched Matching and Didi Chuxing’s Learning and Planning approach, we develop a novel algorithm adapting the latter to a fleet of Autos – self-driving ride-hailing vehicles – and Garages – specialized hubs for storage and refueling. By compiling driver-rider matching, idling, storage, refueling, and redistribution decisions in one unifying framework, we enable a system-wide optimization approach for self-driving ride-hailing previously unseen in the literature. In contrast with existing literature that labeled driverless taxis as economically infeasible, we found that substituting Autos for conventionally driven vehicles stands to increase platform earnings between 90.4% and 99.0% even while bearing the cost of vehicle financing, licensing, maintenance, cleaning, fuel, and oversight previously paid by contracted drivers. Along with increased earnings, the substitution can lower pickup times, improve match rates, and decrease emissions. By adjusting parameters, it is possible to incentivize matching decisions that lower traffic congestion or street parking usage. Our sensitivity analysis indicates that these results are resilient to changing circumstances including high gas prices and policy regulations. We conclude by stating avenues for further improving the model and recommending that city governments take a proactive role in self-driving ride-hailing transitions in order to capitalize on the benefits of the technology while effectively mitigating its harms.
Morris, Jack, "Learning & Planning for Self-Driving Ride-Hailing Fleets" (2020). Undergraduate Honors Theses. William & Mary. Paper 1496.
On-Campus Access Only