6-DOF UR3 Robot Manipulation Based on Deep Learning

According to the traditional calculation method, the UR3 robot needs to conduct complex processing of the image and solve the kinematic model. It is prone to deviation and cannot meet the accurate grasp. This paper introduces a 6-degree-of-freedom (DOF) UR3 robot manipulation method based on deep learning. First, in this paper, the 3D model and the forward kinematic model were established for the UR3 robot. The pose of the target object is calculated by using the pre-trained model with Visual VGG16 as the backbone network, combined with the Perspective-n-Point (PnP) algorithm. Then build the Unity-ROS system, in which the Unity subsystem realizes the simulated grasping work of the UR3 model, and the ROS subsystem realizes the prediction work of the orientations. The feasibility of the method is verified by MATLAB simulation. This paper provides a new, fast and accurate pose estimation method for UR3 robot grasping, which has important practical application value.