The Research On Robotic Grasping Planning Based On 3D Vision And Geometric Features

With the advancement of science and technology,the wide application of industrial robots has brought tremendous changes to human life.It can replace humans in harsh or complicated work and help people complete their work tasks more efficiently.In a factory environment,the most common operation is the process of identification-grasp-place.In order to achieve this repeated process,the position of the target object must be calculated and the posture of the robot hand must be obtained.Therefore,in the actual industrial environment,it is an important challenge in the field of robots to quickly obtain an effective grasping position of an object and perform a grasping operation.Aiming at the problem of grasping planning of industrial robots in real life,the purpose is to obtain a set of grasping points on a target object placed in an arbitrary pose.In this paper,a method of grasping point calculation based on 3D point cloud and its geometric features is proposed.By analyzing the geometric shape factors of the object,the corresponding grasping evaluation function is formulated and verified by the experimental platform.The specific research content is as follows:Research on the calculation method of grasping points based on 3D point cloud.This paper uses a single 3D point cloud to calculate the effective grasping points of unknown objects in the scene.This method uses the point cloud library(PCL)to segment the objects in the scene to obtain the geometric shape attributes of a single target object point cloud.Based on the camera perspective,two grasping strategies were developed,and based on the consideration of the object's geometric characteristics,six factors that affect the stability of object grasping were proposed,and the normalized grasping evaluation function was constructed.The results show that this method can be used to calculate the grasping points of unknown objects,and the calculation speed is faster.A robot grasping experiment platform based on ROS was built.First the RealSense D435 camera calibration and hand-eye calibration were completed.Then a ROS+MoveIt! simulation experiment environment was built under Ubuntu environment to complete the posture output of grasping contact points.Finally,based on the calculated position and pose of the grasping point,the grasping operation of the unknown object was realized,and the reliability and operability of the robot vision grasping system were verified.