108學年度：深度學習技術在影像引導式智慧機器手臂的研究
計畫名稱：深度學習技術在影像引導式智慧機器手臂的研究
執行起迄：2019/08/01~2020/10/31
總核定金額：867,000元
中文摘要：人工智慧技術在各應用領域中的最後一里路，是物理世界中的實體移動。為此，能夠主動或協助人類進行實體物理移動的智慧型機器人，扮演極為重要的角色。未來對於智慧型機器人實體應用時，具備影像辨識、理解、歸納，進而自主規劃拾取、傳遞等功能的機械手臂，是主要核心技術。然而，面對物件的多樣性與不同應用場合，傳統機械手臂的驅動方式必須大幅改進。本計畫探討如何將深度學習技術與實體多軸機械手臂整合，以進行智慧自動辨識物件與動作規劃，達到實現AI最後一里路的實際應用。針對上述議題，本計劃主要工作以深度學習技術設計多軸機械手臂的多角度景深機器視覺，與辨識不規則物件，並決定該物件的夾取方式。內容包含深度學習多角度物件融合、多角度物件形狀模型建立、物件對應拾取物理位置自動標記等相關技術開發。計畫已經完成原訂規劃項目，並延續執行第二階段計畫---深度學習技術在影像引導式智慧機器手臂的研究 II（109-2221-E-390 -016）
英文摘要：Moving physical objects is the last mile of all intelligent applications. Hence, a smart and autonomous robotic arm that can help move objects for humans is the essential role in this scenario. Such a robotic arm needs to detect, recognize, understand the objects to be gripped, and decide how to drive all servos to complete the movement mission. Traditional methods cannot handle the movement control of robotic arms for grabbing the various types of objects. The project explores how deep learning techniques can be integrated with multi-axis robotic arms to recognize multiple types of objects and the decision of servos control for movement. In this project, various objects associated with their gripping positions are modeled by deep convolutionary neural networks. Multiple image sources are fused. The arm is expected to detect, recognize, and decide the gripping position even for the objects that are not seen. The project has achieved the aforementioned tasks and motivates an advanced plan extending these methods -- MOST 109-2221-E-390 -016.
執行起迄：2019/08/01~2020/10/31
總核定金額：867,000元
中文摘要：人工智慧技術在各應用領域中的最後一里路，是物理世界中的實體移動。為此，能夠主動或協助人類進行實體物理移動的智慧型機器人，扮演極為重要的角色。未來對於智慧型機器人實體應用時，具備影像辨識、理解、歸納，進而自主規劃拾取、傳遞等功能的機械手臂，是主要核心技術。然而，面對物件的多樣性與不同應用場合，傳統機械手臂的驅動方式必須大幅改進。本計畫探討如何將深度學習技術與實體多軸機械手臂整合，以進行智慧自動辨識物件與動作規劃，達到實現AI最後一里路的實際應用。針對上述議題，本計劃主要工作以深度學習技術設計多軸機械手臂的多角度景深機器視覺，與辨識不規則物件，並決定該物件的夾取方式。內容包含深度學習多角度物件融合、多角度物件形狀模型建立、物件對應拾取物理位置自動標記等相關技術開發。計畫已經完成原訂規劃項目，並延續執行第二階段計畫---深度學習技術在影像引導式智慧機器手臂的研究 II（109-2221-E-390 -016）
英文摘要：Moving physical objects is the last mile of all intelligent applications. Hence, a smart and autonomous robotic arm that can help move objects for humans is the essential role in this scenario. Such a robotic arm needs to detect, recognize, understand the objects to be gripped, and decide how to drive all servos to complete the movement mission. Traditional methods cannot handle the movement control of robotic arms for grabbing the various types of objects. The project explores how deep learning techniques can be integrated with multi-axis robotic arms to recognize multiple types of objects and the decision of servos control for movement. In this project, various objects associated with their gripping positions are modeled by deep convolutionary neural networks. Multiple image sources are fused. The arm is expected to detect, recognize, and decide the gripping position even for the objects that are not seen. The project has achieved the aforementioned tasks and motivates an advanced plan extending these methods -- MOST 109-2221-E-390 -016.
Click Num: