Demand-responsive rebalancing zone generation for reinforcement learning-based on-demand mobility

Abstract

Enabling Ride-sharing (RS) in Mobility-on-demand (MoD) systems allows reduction in vehicle fleet size while preserving the level of service. This, however, requires an efficient vehicle to request assignment, and a vehicle rebalancing strategy, which counteracts the uneven geographical spread of demand and relocates unoccupied vehicles to the areas of higher demand. Existing research into rebalancing generally divides the coverage area into predefined geographical zones. Division is done statically, at design-time, impeding adaptivity to evolving demand patterns. To enable more accurate dynamic rebalancing, this paper proposes a Dynamic Demand-Responsive Rebalancer (D2R2) for RS systems. D2R2 uses Expectation-Maximization (EM) technique to recalculate zones at each decision step based on current demand. We integrate D2R2 with a Deep Reinforcement Learning multi-agent MoD system consisting of 200 vehicles serving 10,000 trips from New York taxi dataset. Results show a more fair workload division across the fleet when compared to static pre-defined equiprobable zones.