金型空間内の不良成型品ピッキングのための画像認識を用いたロボットシステムの検討

抄録

In this paper, we report the result on basic research using a pick and place robot for removing resin mold articles that have unfortunately fallen on one of molds in an injection molding machine in which multiple molds are arranged for mass production. The position and orientation of target workpieces are essential for successfully conducting a picking task, so that they are tried to be estimated using our developed software for image processing and design of convolutional neural network (CNN). The CNN is one of typical AI technologies specialized in image recognition. The CNN used in experiments is designed through AlexNet-based transfer learning, for estimating the orientations of workpices. The effectiveness of this system is evaluated through pick and place experiments using an articulated robot named DobotMagician.