Explainable reinforcement learning in job-shop scheduling: A systematic literature review

Abstract

Due to the dynamic nature of modern production systems, production planning and control tasks, such as job-shop scheduling, are often solved using reinforcement learning. However, the black-box nature of reinforcement learning hinders the understanding of its decision-making. Explainable reinforcement learning provides explanations of the decision-making, thus increasing trust and transparency, enabling the real-world adoption of such systems. The aim of this work is to establish a foundation for future research on the explainability of reinforcement learning in scheduling a production system, by identifying the state-of-the-art. Therefore, a systematic literature review has been conducted to identify the current knowledge frontier and gaps in knowledge. From our literature review it can be deduced that research on job-shop scheduling using reinforcement learning seldom addresses the explainability of the decision-making. We identified few explainable reinforcement learning techniques in this field and propose that further comprehensive experimental analysis is still required.