Planning and Operation of Ride-Hailing Networks with a Mixture of Level-4 Autonomous Vehicles and For-Hire Human Drivers

This paper investigates the joint planning and operation of a ride-hailing network with a mixture of level-4 autonomous vehicles (AVs) and for-hire human drivers. Distinct from most existing works that focus on fully autonomous vehicles, we envision a more foreseeable future where ride-hailing platform utilizes a fleet of level-4 AVs to provide mobility-on-demand services in selected areas of the city, complemented by a group of for-hire drivers who can pick up or drop off passengers outside of the AV’s service area. The platform not only makes planning decisions regarding the selection of service area of AVs, but also makes operational decisions regarding the ride fares, driver compensations, fleet size, and vehicle relocation strategies, etc., while interacting with passengers and for-hire human drivers on the transportation network. The overall profit-maximization problem of the platform is formulated as a mixed-integer nonlinear program, which is highly non-convex and difficult to address. To tackle this challenge, we developed a decomposition-based algorithm that can compute a near-optimal solution for the optimization problem, while offering a theoretical upper bound on the gap between the derived solution and the unknown globally optimal solution. The proposed method is validated in a numerical study for San Francisco. We show that activating level-4 AVs in selected areas (instead of over the entire network) can improve the platform profit by up to 20%. We also observe that as the cost of AV infrastructures reduces, the spatial diffusion of AV services will follow an interesting pattern, where the platform will first activate all high-demand zones for AV services simultaneously, and then progressively expand AV services to other areas one zone after another as a result of the complex trade-off between demand levels, activation cost, and network connectivity. We further evaluate the impacts of AV diffusion on human drivers and passengers and demonstrate that (a) human drivers will first be incentivized to relocate to lower-demand areas and then be forced to leave the TNC market if the cost of AV infrastructure continues to decline; (b) passengers in remote areas will surprisingly experience higher trip costs even though the cost of AV decreases, particularly after AV is activated in urban areas but before it is activated in remote areas. This finding reveals the unequal distribution of benefits in the autonomous ride-hailing network, underscoring the need for additional regulatory involvement.