Research on the Optimization of the Feeding Mode of the Pick-and-Place Robot Based on Supervised Learning and Stereo Vision

This study is to determine the optimal feeding mode for the tray input by stereo vision characteristics and supervised learning to identify the tilted and disordered workpieces. The 3D image correction of the Moire is applied to the calibration at different heights in the tray. The support vector machine method as supervised learning mainly leads to carrying out the optimal feeding mode for various input directions to object characteristics. The optimization will have different scores (identify the contour of the workpiece) and each feeding method can be analyzed to determine the best feeding mode according to the position, posture and contour. The method of depth scan calculation demonstrated that the classification on grouping SVM with 9 slices scan can resolve the depth resolution of 0.37mm. Searching the priority pick-and-place target can be determined by area size priority and area continuity priority. The goal is to find the object for the top and largest area every time. The optimization has different scores (area size and area continuity) to identify the contour of the workpieces and each feeding method can be analyzed to determine the best feeding mode (the left path).