A Reinforcement Learning Platform for Small and Medium-sized Enterprises in Logistics

Abstract

While real-time data and sophisticated Reinforcement Learning (RL) approaches are emerging, logistic organizations, in particular Small and Medium-size d Enterprises (SMEs), lack the tools and expertise to effectively identify whether (parts of) their business processes are suitable for using RL and adopt these approaches in their daily practice. This paper presents the results of our efforts to design, develop, test and implement a RL-based decision support platform based on the Open Trip Model (OTM) for the logistics industry . The main contribution of this paper is a potentially generalizable platform architecture and instantiation of the following main functional components: 1) a graphical user interface to access platform services, 2) APIs to automate the data collection based on the OTM, 3) an OTM compliant data storage, 4) a repository and tool for testing multiple algorithms and perform (hyper)parameter tuning, and 5) infrastructure provisioning to run and monitor agents. The platform is complemented with RL guidelines to transfer the platform, knowledge, and practices to SMEs. The platform architecture is validated with a panel of experts and demonstrated in use in a case study at a logistics services provider. The platform demonstration and case study contribute to increasing awareness of potential applications of RL in the logistics industry. The platform provides a foundation for empirical research and experimental development of RL approaches in logistics. Future research will focus on alternative and hybrid approaches, federated learning, and incorporating data sharing concepts as part of the envisioned federated data sharing infrastructure for the Dutch logistics industry.