基於大型語言模型之智慧機器人製造單元助理

近年來，大型語言模型（LLMs）在文本生成、機器翻譯與教育輔助等多元應用中展現出卓越的表現。然而，當其應用於現實環境，特別是涉及實體互動場域時，效能常受限制。主要的挑戰包括：模型無法即時取得現場資料、對實務故障排除流程理解不足，以及對機器人動作中隱含的複雜性掌握有限。為解決上述問題，本研究提出一種基於Server-Client架構、整合多代理大型語言模型的控制框架，以提升機器人製造工作站之穩定性、擴展性與作業效率。開發智慧機器人製造單元AI助理，採用TCP（Transmission Control Protocol）通訊協定並以JSON格式進行訊息交換，實現跨平台整合。本系統相較於傳統人工處理方式，展現出更佳效能，特別在以下三項關鍵任務中表現亮眼：自然語言生成機器人控制程式、文字轉換為SQL（Structured Query Language）查詢指令，以及基於RAG（Retrieval-Augmented Generation）技術的語意型故障診斷。實驗結果顯示，本系統不僅提升了機器人控制的準確性與生產資料查詢能力，更將故障處理時間縮短47.6%，顯著強化維修效率與決策支援能力。

Large language models (LLMs) have recently demonstrated impressive capabilities in diverse applications such as text generation, machine translation, and educational assistance. However, when applied to real-world scenarios, particularly those involving physical interactions, their performance is often limited. This limitation is primarily due to their lack of access to on-site data, insufficient understanding of practical troubleshooting procedures, and limited comprehension of the complexities inherent in robotic actions. Hence, this article proposes a Server-Client control framework augmented with a multi-agent LLM to enhance the robustness, scalability, and efficiency of robotic manufacturing workcells. Our system develops an AI assistant that outperforms traditional manual methods by using a centralized TCP (Transmission Control Protocol)-based protocol with JSON messages for cross-platform integration. It excels in three tasks: robot code generation from natural language, text-to-SQL (Structured Query Language, SQL) query, and fault diagnosis retrieval using a RAG (Retrieval-Augmented Generation)-based semantic search. Experiments show improved robot control, accurate production data queries, and a 47.6% reduction in fault handling time, boosting maintenance efficiency and decision support.