Sparsity-Aware Grouped Reinforcement Learning for Designated Driver Dispatch

Abstract

Designated driving service is a fast-growing market that provides drivers to transport customers in their own cars. The main technical challenge in this business is the design of driver dispatch due to slow driver movement and sparse orders. To address these challenges, this paper proposes Reinforcement Learning for Designated Driver Dispatch (RLD3). Our algorithm considers group-sharing structures and frequent rewards with heterogeneous costs to achieve a trade-off between heterogeneity, sparsity, and scalability. Additionally, our algorithm addresses long-term agent cross-effects through window-lasting policy ensembles. We also implement an environment simulator to train and evaluate our algorithm using real-world data. Extensive experiments demonstrate that our algorithm achieves superior performance compared to existing Deep Reinforcement Learning (DRL) and optimization methods.