Achieving Stable and Optimal Passenger-Driver Matching in Ride-Sharing System

Abstract

Ride-sharing systems enable individual car owners with idle time to provide commercial taxi-like services via an online platform. By crowdsourcing a large population of individual car owners, it can provide more flexible services with a lower serving cost, comparing with the traditional taxi system. Due to the autonomous nature of car owners (drivers), a decentralized driver dispatching algorithm that can achieve a stable (self-motivated) and optimal passenger-driver matching is highly desired for a ride-sharing system. In this paper, we will study such a driver dispatching algorithm systematically. We first show that the optimal passenger-driver matching achieved by the centralized driver dispatching algorithm is often not stable, in the sense that some drivers and passengers may break with their matched partners and form new matching pairs. To this end, we introduce a virtual order fee on each passenger (which the platform will charge the drives who want to serve the passenger) to motivate the behaviors of drivers. Specifically, we propose a novel auction-based decentralized driver dispatching algorithm, where each driver proposes the most profitable passenger that he wants to serve, considering the potential profit that he can achieve and the order fee that he needs to pay from/to serving each passenger. The virtual order fee on a passenger will be gradually increased when multiple drivers want to serve the passenger, until there exists only one driver who is willing to serve. We analytically show that such a decentralized driver dispatching algorithm will converge to an equilibrium (stable) outcome, which achieves the optimal passenger-driver matching (i.e., that maximizes the social income of the whole system). Simulation results further show how the converging speed and the achieved social income change with the system parameters such as the step size of order fee increasement. Moreover, it is easy to implement the proposed distributed algorithm in a practical system.