6D Robotic Grasping System using Convolutional Neural Networks and Adaptive Artificial Potential Fields with Orientation Control

Abstract

In this paper, a new robotic grasping pipeline is implemented in an additive manufacturing unit to autonomously pick printed objects in a 3D printer, while avoiding obstacles in the path to the grasp pose. This is accomplished through the integration of a 6D grasp generator (GraspNet), an instance segmentation method (Mask R-CNN), a point cloud collision check system, and a path planning technique with orientation control based on the Adaptive Artificial Potential Field algorithm. The Robot Operating System (ROS) framework and a collaborative robot manipulator UR5 are used to validate the proposed method in a task of picking an object from a 3D printer tray. The performance analysis of the proposed system is presented through simulation with objects of complex geometry using Webots.