本論文的任務目標在於藉由導入影像視覺系統，開發一種不需要使用CAD模型匹配的辨識夾取策略，其中整合了二維與三維點雲資料，透過深度學習網路檢測目標物以及對目標物點雲進行最佳六自由度(6 Degrees of Freedom, 6DoF)夾取姿態估計，並在機構限制下利用逆向運動學控制六軸機械手臂對四種不同的目標物在工作區域內隨意平放的情況下執行夾取與分類任務，夾取與分類任務的成功率為86%，成果顯示本論文確實能成功建立一套物件辨識與夾取分類系統。;Robot operating system (ROS) is used to develop a software system under the Ubuntu 18.04 version of Linux environment in this study. The industrial IPC, the robot arm, the binocular structured light camera and the grippers are integrated by ROS distributed architecture and peer-to-peer network, and all information and data collected can be transferred to them as well. Therefore, the collaborative design is used to realize the integrated software and hardware.
The main purpose of this paper is to develop a recognition and gripping strategy that does not require CAD model matching through the introduction of the depth image vision system, to identify the target through the neural network and automatically generate a six-degree-offreedom (6DoF) gripping pose for the object, and the success rate of gripping and sorting is 86% under the condition that the six-axis robot arm is controlled to place four different objects in the working area at will under the limitation of the mechanism. The results show that this paper can indeed successfully establish a set of object recognition and gripping system.