Title: Scalable Ride-Sourcing Vehicle Rebalancing with Service Accessibility Guarantee: A Constrained Mean-Field Reinforcement Learning Approach

Abstract:

The rise of ride-sourcing platforms like Uber and Lyft has fundamentally transformed urban transit, providing flexible, on-demand mobility through smartphone applications. However, these services face substantial operational hurdles, most notably vehicle rebalancing—the strategic repositioning of a fleet to correct spatiotemporal imbalances between supply and demand. Poor rebalancing strategies lead to extended passenger wait times, suboptimal vehicle usage, and significant equity concerns, including unequal service distribution and income disparities among drivers.

To address these challenges, we propose continuous-state Mean-Field Control (MFC) and Mean-Field Reinforcement Learning (MFRL) models that utilize continuous repositioning actions. By modeling each vehicle’s behavior through its interaction with the overall vehicle distribution rather than individual agents, MFC and MFRL provide scalable solutions. This approach alleviates the curse of dimensionality associated with the number of agents, facilitating coordination across large fleets with substantially lower computational complexity. Furthermore, it removes the necessity for model retraining when fleet sizes fluctuate.

To guarantee equitable service access across different geographic areas, we incorporate an accessibility constraint into our models. This allows for the derivation of rebalancing policies that balance high rider demand fulfillment with fair vehicle supply coverage. Data-driven simulations conducted in Shenzhen validate the efficiency and robustness of our method. Notably, the approach scales to fleets comprising tens of thousands of vehicles, with training times similar to those of linear programming rebalancing. Our policies effectively navigate the efficiency-equity Pareto front, surpassing conventional benchmarks in critical metrics such as fleet utilization, fulfilled requests, and pickup distance, all while maintaining equitable service access.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC