隨智慧製造發展，多移動機器人分散式協同運作已成提升物流彈性的核 心，然而嵌入式微控制器執行 micro-ROS 通訊時，常因控制演算法與通訊協 定棧於單一核心混合排程，引發任務搶佔並導致端到端時延產生隨機突波。為 解決此瓶頸，本研究提出一套基於 MIAT 方法論之模型導向 micro-ROS 分散 式系統重建流程，並引入大型語言模型（LLM）作為自動軟體合成引擎。本方 法利用 IDEF0 模型解構任務流與資料流邊界，透過 GRAFCET 建立並行行為 模型與程式碼映射規則，並將模型約束轉為結構化提示詞，驅動 LLM 合成具 備雙核心實體隔離與 FreeRTOS 執行緒綁定之 C 語言原始碼。定量實驗證 實，單節點組態下平均單向時延由 17.49 毫秒降至 9.94 毫秒；於 2 台、5 台 與 10 台設備擴展性測試下，總吞吐量呈線性增長且維持 90.9% 以上之接收飽 和度。本研究貢獻在於融合形式化模型與生成式 AI 實現設計自動化，消除高 階設計與低階實作落差，為多機器人系統提供具確定性的底層通信保障。;With the advancement of smart manufacturing， the distributed collaboration of multiple Robots has become pivotal for enhancing logistics flexibility. However， conventional microcontrollers executing distributed micro-ROS communication often suffer from task preemption due to the hybrid scheduling of control algorithms and communication stacks on a single core， leading to random spikes in end-to-end latency. To address this bottleneck， this study proposes a MIAT-based model-driven micro-ROS reconstruction flow and introduces a Large Language Model (LLM) as an automated software synthesis engine. The proposed framework utilizes IDEF0 models to decouple the boundaries between control tasks and communication data streams， and applies GRAFCET to construct parallel behavior models along with code mapping rules. These high-level model constraints are transformed into structured prompts， driving the LLM to synthesize C code with FreeRTOS thread bindings and hardware-level dual-core isolation. Experimental results demonstrate that under a single-node configuration， the proposed framework reduces the average one-way latency from 17.49 ms to 9.94 ms. In scalability stress tests with 2， 5， and 10 devices， the total system throughput exhibits linear growth， maintaining a packet delivery saturation rate above 90.9%. The contribution of this research lies in the integration of formal models and generative AI to achieve efficient design automation， bridging the gap between high-level designs and low-level source code while guaranteeing deterministic distributed communication for multi-robot systems.
