Taxonomy of LLM Integration in MARL for Smart Manufacturing
August 10, 2026
This architecture explores four LLM attachment points within Multi-Agent Reinforcement Learning (MARL) systems: policy, reward design, agent communication, and hierarchical planning. It categorizes LLM utility based on engineering maturity and formal performance guarantees in Dec-POMDP environments.
HOW THIS AFFECTS YOU
●
builderYou can use this taxonomy to decide where to plug LLMs into existing multi-agent control systems.
●
researcherThe framework provides a structured way to evaluate LLM-augmented reinforcement learning in partially observable environments.