An ArUco-based Grasping Point Detection System.

This study develops a grasping point detection method for robotic arms. The proposed method, named VGG16-GPDNet, can assist robotic arms in estimating objects’ pose and position in space. VGG16-GPDNet, an improved version of VGG16, is used to train deep learning models that input images of target objects and output object grasping points. It is generally difficult to calibrate the grasping point values of an object; therefore, the ArUco marker is used as a tool for calibrating object grasping points without requiring complex computation. The experiment in this study verified that applying model-predicted grasping points in the arm grasping test yielded a success rate of 92%. The proposed method enables robotic arms to grasp and transfer objects in real-life environments.