Autonomous Vehicles for Ride-Hailing

Abstract

Problem definition: We consider a setting in which a ride-hailing platform operates a mixed fleet of conventional vehicles (CVs) and autonomous vehicles (AVs) over locations distributed spatially. The CVs are operated by human drivers who make independent decisions about whether to work for the platform and where to position themselves when they become idle. The AVs are under the control of the platform. The platform decides on the wage it pays the drivers, the size of the AV fleet and how AVs are positioned spatially when they are idle. The platform can also make decisions on how much demand to accept for each pair of origins and destinations and whether to prioritize AVs or CVs in assigning vehicles to customer requests. Methodology: We use a fluid model to characterize the optimal decisions of the platform in equilibrium and contrast those with the optimal decisions in the absence of AVs. We study outcomes in equilibrium in terms of platform profit, customer service level, driver welfare, and driver productivity. Results: Among our findings, we show that the platform, whenever possible, would deploy the AVs in such a way as to reduce repositioning by CVs from the low demand location to the high demand location (inducing such repositioning by drivers is costly for the platform). The presence of AVs can also eliminate driver incentives that would otherwise force the platform to reject demand from the low demand location to force drivers to relocate to the high demand location. Even though demand is no longer exclusively fulfilled by CVs, we show that the income of drivers may not necessarily be harmed, and that drivers may not necessarily experience more idleness or empty travel. Moreover, we find that drivers can be strictly better off if the platform prioritizes AVs in assigning customer requests to vehicles. Managerial implications: Our results uncover important ways the introduction of AVs affects the operation of a ride-hailing platform and highlight the nuanced impact of AVs on human drivers and customers. Our results are potentially useful to policy makers in deciding on regulatory interventions that can induce more socially desirable outcomes with the introduction of AVs.

Keywords: autonomous vehicles, ride-hailing, equilibrium fluid model, driver welfare, vehicle repositioning, admission control, assignment priorities