Abstract
Large-scale online ride-sharing platforms have substantially transformed ourlives by reallocating transportation resources to alleviate traffic congestionand promote transportation efficiency. An efficient fleet management strategynot only can significantly improve the utilization of transportation resourcesbut also increase the revenue and customer satisfaction. It is a challengingtask to design an effective fleet management strategy that can adapt to anenvironment involving complex dynamics between demand and supply. Existingstudies usually work on a simplified problem setting that can hardly capturethe complicated stochastic demand-supply variations in high-dimensional space.In this paper we propose to tackle the large-scale fleet management problemusing reinforcement learning, and propose a contextual multi-agentreinforcement learning framework including two concrete algorithms, namelycontextual deep Q-learning and contextual multi-agent actor-critic, to achieveexplicit coordination among a large number of agents adaptive to differentcontexts. We show significant improvements of the proposed framework overstate-of-the-art approaches through extensive empirical studies.