Robotic System Integrated with Object Classification and Gripping Pose Estimation

This research leverages the Robot Operating System (ROS) to develop a software system that integrates an industrial computer, a six-axis robotic arm, a depth camera, and an adaptive gripper through ROS's distributed architecture. The objective is to devise a recognition and grasping strategy without CAD model matching by fusing 2D and 3D point cloud data. A deep learning network identifies target objects and determines optimal six degrees of freedom (6DoF) grasp poses. The robotic arm is controlled using inverse kinematics to grasp and sort four objects randomly positioned in the workspace, resulting in an 86% success rate and showcasing the system's efficacy.