AMARL Centered Reference Architecture for Large Language Model Augmentation in Smart Manufacturing
The hosts discuss a paper proposing a reference architecture for adding large language models (LLMs) to multi-agent reinforcement learning (MARL) in smart manufacturing. They cover four LLM attachment points, a three-layer architecture, a decision framework, and evidence showing most studies remain at simulation level. They conclude LLMs should augment, not replace, MARL for safety-critical control.