Research On Grasping Method Of Irregular Parts Based On Improved GG-CNN Algorithm

Robot grasping technology for targets with known shape rules has been very mature,but it cannot meet the needs of industrial intelligence for intelligent robot technology.With the development of sensor technology and the continuous improvement of deep learning algorithms,it is possible to grasp unknown irregular parts in unstructured real scenes by using visual inspection methods based on deep learning.This paper studies the detection algorithm based on deep learning to locate and grasp unknown irregular parts in industry and improve the intelligent operation level of robots.The irregular part grasping is modeled and simulated in the Pybullet simulation platform,and then an experimental platform is built for verification.Firstly,Matlab software is used to calibrate the depth camera used,and the environmental information and object depth information in the grasping scene are obtained by using the depth camera.The depth image was processed and analyzed by the image matching algorithm and the depth information repair algorithm,so as to improve the expression ability of object depth information.Secondly,in order to improve the labeling efficiency of the grasp pose,the point-line grasp model is improved by converting the pixel-level labeling into region-level labeling,which reduces the time required to label the target in the dataset and optimizes the conflict problem of grasp angles.Thirdly,in order to realize that the manipulator can grasp unknown irregular parts in unstructured grasping scenes,the GG-CNN algorithm is used,and its network structure and loss function are improved,which improves the generalization ability of the GG-CNN network for grasping small target objects and objects with rich depth information.The prediction accuracy in the data set is about 8.5% higher than that of the original algorithm.Finally,the kinematic analysis and workspace analysis of the manipulator were carried out,and the hand-eye calibration of the depth camera and the manipulator was carried out to improve the grasping accuracy.The manipulator model and simulation experiment platform were built in Pybullet simulation system,and the target grabbing capability based on GG-CNN network was analyzed.The experimental operation further verifies the feasibility of the proposed method for grasping irregular parts.