"Truly" Empty Vehicle Repositioning and Fleet-Sizing: Optimal Management of an Autonomous Taxi System in New Jersey on a Typical Weekday

Abstract

Transportation is central to our everyday lives and exists to increase the utility of all. Since the turn of the twentieth century, innovators have drastically revolutionized the field. Yet still today, inefficiencies exist in our daily routines in terms of cost, safety, mobility, comfort, convenience, and environmental impact. Already, work has begun to analyze a potential system of autonomous taxis for the state of New Jersey from the perspective of local demand on a typical day. However, the practical issues of sizing a finite fleet of autonomous taxis and repositioning the vehicles to best meet stochastic and realized travel demand need to be addressed. As fully autonomous vehicles more closely approach mainstream use, the gravity of these issues magnifies significantly. This thesis first defines properties of a stochastic queueing network and outlines the significant characteristics of an autonomous taxi system for New Jersey with an infinite fleet of vehicles under different parameter sets. The understanding of ride sharing, vehicle departures, and vehicle arrivals in the infinite fleet case forms a necessary starting point for work on empty vehicle repositioning and fleet-sizing. The thesis outlines a series of naïve empty vehicle repositioning and fleet-sizing policies before providing a formal and detailed mathematical model for the empty vehicle repositioning problem specific to an autonomous taxi system. It then provides policies for empty vehicle repositioning and fleet-sizing based on the model. However, these policies share a high level of complexity and require extremely large dimensionality. The thesis follows up with a series of simpler, alternative policies that require more manageable levels of complexity, and presents some preliminary results on a small but populous subset of New Jersey. Finally, this thesis presents techniques for convenient data visualization through a detailed web application. Optimal management of even a small-scale autonomous taxi system on a typical day is extremely challenging. The massive, continuous data streams that come with a larger-scale system present further computational difficulties. Research never stops; I hope that this thesis serves as a foundation for further work of my own and of others. This is one small step among many larger ones. Together, we can make autonomous taxi systems a reality.