Robust Vehicle Rebalancing with Deep Uncertainty in Autonomous Mobility-on-Demand Systems

TL;DR

This paper presents DURO, a deep neural network-based robust optimization framework for vehicle rebalancing in autonomous mobility-on-demand systems, effectively managing demand uncertainty and outperforming traditional models in accuracy and efficiency.

Contribution

Introduces DURO, a novel deep uncertainty robust optimization framework that improves vehicle rebalancing in AMoD systems under demand uncertainty using neural networks.

Findings

DURO outperforms deterministic models in demand forecasting accuracy.

DURO achieves comparable results to DRO with better computational efficiency.

Real-world NYC data validates DURO's effectiveness in managing demand uncertainty.

Abstract

Autonomous Mobility-on-Demand (AMoD) services offer an opportunity for improving passenger service while reducing pollution and energy consumption through effective vehicle coordination. A primary challenge in the autonomous fleets coordination is to tackle the inherent issue of supply-demand imbalance. A key strategy in resolving this is vehicle rebalancing, strategically directing idle vehicles to areas with anticipated future demand. Traditional research focuses on deterministic optimization using specific demand forecasts, but the unpredictable nature of demand calls for methods that can manage this uncertainty. This paper introduces the Deep Uncertainty Robust Optimization (DURO), a framework specifically designed for vehicle rebalancing in AMoD systems amidst uncertain demand based on neural networks for robust optimization. DURO forecasts demand uncertainty intervals using a deep neural network, which are then integrated into a robust optimization model. We assess DURO against various established models, including deterministic optimization with refined demand forecasts and Distributionally Robust Optimization (DRO). Based on real-world data from New York City (NYC), our findings show that DURO surpasses traditional deterministic models in accuracy and is on par with DRO, but with superior computational efficiency. The DURO framework is a promising approach for vehicle rebalancing in AMoD systems that is proven to be effective in managing demand uncertainty, competitive in performance, and more computationally efficient than other optimization models.