Graph Neural Network Reinforcement Learning for Autonomous Mobility-on-Demand Systems
This addresses the challenge of efficient fleet management in autonomous transportation systems, offering incremental improvements in policy transferability and scalability.
The paper tackles the control problem for autonomous mobility-on-demand systems by proposing a deep reinforcement learning framework using graph neural networks, resulting in policies that are more transferable, generalizable, and scalable than other approaches, with demonstrated zero-shot transfer capabilities in tasks like inter-city generalization and urban topology adaptation.
Autonomous mobility-on-demand (AMoD) systems represent a rapidly developing mode of transportation wherein travel requests are dynamically handled by a coordinated fleet of robotic, self-driving vehicles. Given a graph representation of the transportation network - one where, for example, nodes represent areas of the city, and edges the connectivity between them - we argue that the AMoD control problem is naturally cast as a node-wise decision-making problem. In this paper, we propose a deep reinforcement learning framework to control the rebalancing of AMoD systems through graph neural networks. Crucially, we demonstrate that graph neural networks enable reinforcement learning agents to recover behavior policies that are significantly more transferable, generalizable, and scalable than policies learned through other approaches. Empirically, we show how the learned policies exhibit promising zero-shot transfer capabilities when faced with critical portability tasks such as inter-city generalization, service area expansion, and adaptation to potentially complex urban topologies.