Grasping Detection Algorithm Based On Machine Vision And Grasping System Design

Intelligent robots rely on a variety of technologies to sense and interact with the environment in order to accomplish their operational tasks.Among them,the implementation of target object grasping depends on grasping technology.Grasping technology has a wide range of applications in manufacturing,service,logistics,and medical fields.With the extensive application of Artificial Intelligence(AI)technology in the field of robotics,the degree of robot intelligence is becoming higher and higher.In robotic arm grasping,the target object grasping detection is a key part of the whole robotic arm grasping system function.Most current grasp detection algorithms require the grasped target to be placed in a simple and clean environment for study.While this approach greatly simplifies the target object localization task,it also means that they have limited performance in target object localization in relatively complex environments.In this thesis,based on a comprehensive understanding of the key technologies required to perform robotic arm grasping,a robotic arm grasping system is constructed based on grasp detection technology.The main research work is as follows:1.The target detection problem and the grasp detection problem are jointly studied for the problem of poor target positioning ability of existing grasp detection algorithms in real environments.Based on the target detection algorithm Center Net,a joint model of target detection and grasp detection with shared features is proposed,which can realize both target object detection and grasp detection functions.In addition,with reference to the Cornell Grasping dataset,a self-built dataset containing both target detection frames and grasping frames is used to verify the detection performance of the proposed model in this thesis,which is tested on both the Cornell Grasping dataset and the self-built dataset.2.A grasp detection system containing hardware layer,algorithm layer and user layer is built in ROS(Robot Operating System)environment.For the grasp detection module,the joint model of target detection and grasp detection proposed in this thesis adopted.Based on the in-depth study of the kinematic solver of the robotic arm and the robotic arm motion planning algorithm,the kinematic solver and motion planner of the robotic arm are configured based on Move It.At last,the object detection and grasping experiments were conducted in Gazebo simulation environment and real environment respectively.With average grasping success rate of 90.6% and 89%,which verified the effectiveness of the algorithm and the feasibility of the system.