產業脈動｜從看見到看懂，機器人視覺技術的演進

本文首先說明全球機器人視覺與工業機器視覺市場的成長動能與技術趨勢，接著探討機器人視覺 由2D 平面辨識升級至3D 空間理解的技術脈絡，分析3D 深度相機、光達與毫米波雷達如何在近距離操作、 遠距離導航與複雜環境避障中形成互補。最後說明深度學習與視覺基礎模型如何讓機器人從單純取得影 像，進一步具備理解環境、判斷任務與調整動作的能力。機器人視覺正從任務專用模型走向通用基礎模 型，並與力覺、觸覺動作資訊融合，形成感知行動的循環。隨著具身智慧的發展，最新技術趨勢也開始 研究將視覺、力覺與觸覺整合至統一感知框架之中。

This article first explains the growth drivers and technological trends of the global robot vision and industrial machine vision markets. It then examines the technological trajectory of robot vision as it evolves from 2D image recognition to 3D spatial understanding, highlighting how 3D depth cameras, LiDAR, and millimeter-wave radar complement one another in close-range manipulation, long-range navigation, and obstacle avoidance in complex environments. Finally, the article discusses how deep learning and visual foundation models enable robots to move beyond simply capturing images toward understanding environments, judging tasks, and adjusting actions. Robot vision is shifting from task-specific models to general-purpose foundation models, while also integrating force sensing, tactile sensing, and action information to form a perception-action loop. With the development of embodied intelligence, the latest technological trends are also exploring the integration of vision, force sensing, and tactile sensing into a unified perception framework.