Publication Type

Journal Article

Version

submittedVersion

Publication Date

8-2026

Abstract

Robotic mobile fulfillment systems play a vital role in modern e-commerce, but their efficiencydepends on overcoming a critical challenge: coordinating orders and racks effectively under dynamicorder arrivals, diverse delivery deadlines, and resource sharing among multiple stations. This papertackles the resulting dynamic order and rack scheduling problem and proposes an intelligentscheduling framework centred on deep reinforcement learning. At its core is TriNet-PPO, whichtrains a scheduling policy using TriNet, a tripartite graph neural network encoder tailored to capturethe many-to-many relationships among orders, racks, and items. Furthermore, we introduce IPCSA,an integer programming-based cooperative scheduling algorithm that maps station-level actionpreferences to conflict-free joint decisions in multi-station environments by resolving real-timerack allocation conflicts. Computational experiments show that the proposed framework achievesstronger overall performance than the baselines in dynamic, multi-station environments.

Keywords

Deep reinforcement learning, Dynamic scheduling, E-commerce, Multiple stations, Order and rack scheduling

Discipline

Artificial Intelligence and Robotics | Operations and Supply Chain Management

Research Areas

Intelligent Systems and Optimization

Areas of Excellence

Sustainability

Publication

International Journal of Production Research

First Page

1

Last Page

29

ISSN

0020-7543

Identifier

10.1080/00207543.2026.2713122

Publisher

Taylor and Francis Group

Additional URL

https://doi.org/10.1080/00207543.2026.2713122

Download

Share

COinS