技術專題主編前言｜Physical AI 引領智慧機器人自動化的發展變革

隨著生成式人工智慧與多元感測技術的進步，全球自動化挑戰正由傳統「快、穩、準」的設備自動 化思維，跨越至處理不確定性自動化能力的「物理 AI」與「具身智能」新紀元。本期前言特別指出，次 世代智慧機器人控制架構正重塑「大腦」與「小腦」的融合，例如將傳統控制器架構適當結合模型基礎與 人工智慧數據驅動方法，使其從單純的運動控制，演進為具備任務規劃與歷史加工記憶功能的智慧型控制 器。專輯依循由宏觀戰略至技術落地的脈絡編排，收錄國內頂尖產學研團隊的研究成果，內容涵蓋隱性製 程經驗轉化、全球具身智能供應鏈布局、物理 AI 落地案例實證，以及協作機器人智慧組裝、異質雙臂協 作搬運與無治具組裝實務。此外，後半段技術專文更深入探討無 CAD 模型之生成式視覺智慧研磨、基於 大型視覺語言模型的無地圖自主導航，以及利用低成本壓力感測陣列解耦絕對重量的底層觸覺感知模型。 綜觀全期，各項創新成果精準回應少量多樣、高變異與高依賴經驗判斷等產業應用痛點，全面展現臺灣 在全球智慧機器人產業鏈中的跨感知整合能量與落地潛力。

Driven by advancements in generative AI and multimodal sensing, global automation is shifting from the traditional "speed, stability, and precision" paradigm to a new era of "Physical AI" and "Embodied AI" capable of handling operational uncertainties. This preface highlights how next-generation intelligent robot architectures are reshaping the "brain-cerebellum" integration. By appropriately combining model-based and AI data-driven methods, traditional controllers are evolving from simple motion control into intelligent controllers equipped with task planning and historical processing memory. Structured from macro-strategies to technical implementations, this issue features top domestic research covering the transformation of implicit manufacturing experience, global embodied AI supply chain layouts, and empirical physical AI deployments. Furthermore, it explores practical breakthroughs in collaborative and jigless assembly, heterogeneous dual-arm collaboration, CAD-free generative vision-guided grinding, map-free autonomous navigation via large vision-language models, and low-cost tactile perception models. Collectively, these innovations precisely address industrial pain points such as high-mix lowvolume production, high process variability, and heavy reliance on experience, comprehensively demonstrating Taiwan's cross-perception integration capabilities and commercialization potential within the global intelligent robotics supply chain.