The Gripping Posture Prediction of Eye-in-hand Robotic Arm Using Min-Pnet

Abstract

This study focuses on using RGB-D images and modifying an existing machine learning network architecture to predict the gripping posture of a successfully grasped object. A five-finger(5-Fin) gripper designed to mimic the human palm was tested to demonstrate that it can perform a more delicate mission than many two- or three-finger grippers. Experiments were conducted using the 6-DOF robot arm with the 5-Fin and 2-Fin grippers to perform at least 100 actual machine grasps, and compared to the results of other studies. It was demonstrated that our network could perform as well as a deep network architecture with little training data and omitting steps such as posture evaluation. When combined with the hardware advantages of the 5-Fin gripper, it can produce an automated system with a gripping success rate of over 90%.