Abstract
The emergence and adoption of fully-autonomous vehicles (AVs) is expected to accelerate certain trends already underway in the transportation sector, such as the growth of shared-use mobility and mobility-on-demand services. This paper models an AV mobility service option wherein a fleet of AVs provide direct origin-destination service to travelers that request rides via a mobile application and expect to be picked up within a few minutes. The underlying problem is highly dynamic and stochastic. The solution strategy consists of repeated solution of an integer program referred to as the AV-traveler assignment problem. As the state of the system changes via dynamic traveler requests entering the system, the AV fleet operator re-solves the static AV-traveler assignment problem to assign and reassign AVâ??s to travelers. Given that AV fleets will need to compete with the personal vehicle in terms of cost and quality of service, the authors present and compare several optimization-based strategies to operate an AV fleet with the twin objectives of minimizing costs and maximizing quality of service. To compare the AV fleet operational strategies, the authors perform an extensive computational analysis using an agent-based simulation tool. The results indicate that using optimization-based heuristic strategies rather than simple first-come, first-served (FCFS) heuristics, and incorporating all AVs in the AV-traveler assignment problem (not only currently idle AVs) improves the efficiency of the AV fleet in terms of fleet miles and traveler wait times. The simulation results also indicate that the most-effective AV-traveler assignment strategy results in 6-7% of all fleet miles to be empty for spatially clustered traveler origins and destinations, compared with 11-15% for traveler origins and destinations that are uniformly distributed.