Best paper award in Manufacturing and Design (D&M) track.

IISE 2025 D&M Track Talk

This talk presents a transferable DRL architecture that can be directly applied across diverse discrete manufacturing systems. Specifically, we propose a fully distributed multi-agent framework that categorizes manufacturing resources into distinct classes based on their functions. Each class of resources requires a similar structure of local information for decision-making, allowing each DRL agent to learn control policy for a class of resources rather than an individual resource. A new observation update method ensures that observable local information is sufficient for distributed agents to make collaborative decisions for a class of resources, regardless of their location within the manufacturing system. As manufacturing systems can be viewed as combinations of these resource classes, agents trained under this framework can be combined flexibly and adapt to various manufacturing system configurations.