This paper aims to develop a robotic system that is able to find and remove unwanted molded articles, which fell in a narrow metallic mold space. Currently, this task is being supported by skilled workers. The proposed robotic system has the ability to estimate the orientation of articles using transfer learning-based convolutional neural networks (CNNs). The orientation information is essential and indispensable to realize stable robot picking operations. In addition, pixel-based visual feedback (PBVF) controller is introduced by referring to the center of gravity (COG) position of articles computed by image processing techniques. Hence, it is possible to eliminate the complex calibration between the camera and the robot coordinate systems. The implementation and effectiveness of the pick and place robot are demonstrated, where the conventional calibration of such task is not required.
- Convolutional neural network (CNN)
- Pick and place
- Pixel-based visual feedback
- Transfer learning
ASJC Scopus subject areas
- Biochemistry, Genetics and Molecular Biology(all)
- Artificial Intelligence