Fair Cruising

Publikation: Bidrag til bog/antologi/rapport/proceedingKonferencebidrag i proceedingsForskningpeer review

Ride-hailing systems operate in a two-sided market between passengers and drivers. Such systems manage a fleet of vehicles via two critical operations on the drivers' side: the selection of a route to take when not serving a passenger, or cruising, and the assignment of a customer to a driver, or dispatching, and thereby affect the market equilibrium. The quality of fleet management has been extensively studied with respect to collective profit on the drivers' side, and satisfaction on the passengers' side, yet less with respect to the satisfaction of drivers. In this paper, we propose a maximim criterion of fleet management quality that expresses the fairness among drivers, with a focus on cruising decisions. We find that state-of-the-art cruising solutions based on Reinforcement Learning perform poorly in terms of his fairness objective compared to simple baselines. We adapt these methods based on an enhanced description of the environmental state, and suggest a fairness-oriented combination of cruising and dispatching decisions. Our results show that this adaptation achieves better fairness than state-of-the-art techniques on real-world and synthetic data.
OriginalsprogEngelsk
TitelProceedings of the ACM SIGKDD Conference on Knowledge Discovery and Data Mining 2020
StatusAfsendt - 13 feb. 2020

Se relationer på Aarhus Universitet Citationsformater

ID: 179365958