Defective Article Picking Robot Using Image Processing Technique and Visual Feedback Control

Abstract

The authors are developing a robot system which can remove defective molded articles in narrow metallic mold space which has been manually done by skilled workers. The robot system can estimate the orientation of articles by using a transfer-learning based convolutional neural network (CNN). The orientation information is essential and indispensable to realize stable robotic picking. In addition, a visual feedback (VF) controller is designed by referring the COG position of articles obtained by image processing, so that the complicated calibration task between camera and robot coordinate systems can be eliminated. As a result, the authors propose a smart pick and place robot system which do not require conventional calibration tasks.