The Search for the Sustainable Fleet: Driverless Taxi System Simulations

Abstract

Within the next several years, driverless vehicle technology may begin to rapidly disrupt the transportation industry in the developed world. Transportation needs would be served with unprecedented efficiency and the entire system of vehicle ownership could be redesigned. This thesis develops a model for the simulation of an aTaxi fleet in New York City serving trip requests and autonomously repositioning its vehicles, parametrized by several vehicle and trip constraints. The fleet simulation is evaluated along various performance metrics including average consumer waiting time, average vehicle repositioning distance, and average vehicle occupancy. The objective of the simulation is to test two hypotheses: (H1) there exists a sustainable fleet sufficiently large enough to yield non-diminishing performance metrics over an extended time period and (H2) the Demand Learning Strategy yields better performance metrics than the No Learning Strategy. The experiment finds H1 to be true with high probability, as daily demand trends exhibit periodic similarity. The experiment finds H2 to be false, but suggests alternate learning strategies to be explored in further research. In practice, the analysis in this thesis is directly applicable to fleet system managers who must handle the issue of balancing user utility with system efficiency in order to provide the best level of service possible.