Research And Simulation Of Intelligent Grasping System Of Service Robot Based On Visio

In the current field of intelligent robot development,industrial robot development is relatively mature,the field of service robot is in the ascendant,especially the household service robot for daily life has gradually become a research hotspot in the field of intelligent robot.At present,the number of service robots that can be applied in practice is small,with simple functions and high costs.Intelligent robots that can master precise grasping and other functions are still in the laboratory stage and have not been widely applied in daily life.This topic,focusing on various problems encountered by service robots in realizing intelligent grasping in family life,is committed to building a low-cost and efficient mobile robot intelligent grasping system.The main achievements include:(1)Aiming at the difficult problem of target detection caused by large range and interference of home environment,a set of multi-sensor fusion visual detection mechanisms are constructed.Kinect V2 and Intel Realsense D435 depth camera are used to complete near and far vision detection and obtain more accurate pose information.(2)Aiming at the problem that the existing Line Mod algorithm for template matching has weak recognition ability under occlusion and poor adaptability to complex family environment,a CMRL method based on 7D feature matching is optimized.By introducing the deeper 7D feature vector of feature points to improve the influence proportion of unique features in the process of feature point classification,the features with obvious correlation on the object surface can be separated well,so that the segmented template has more identifiable information and the recognition rate can be improved.Experimental results show that the CMRL method can improve target recognition and pose estimation performance significantly.(3)In this paper,lidar is integrated into the depth camera-based RTAB-Map method,and the feedback information of lidar odometer is used to obtain a more accurate depth map for robot navigation.(4)Aiming at the goal of grasping multiple types of objects efficiently,quickly and stably in the home environment,this paper constructed robot grasping methods for known,similar and unknown objects respectively.The Dg GNet algorithm is used to construct the grasping experience database,and the target detection system is used to obtain the target information to quickly complete the generation of grasping strategies for most of the known and similar objects in home life.To solve the problem of grasping unknown objects,this paper optimized a deep grasping detection network based on the Res Net framework.Grasping area and antipodal grasping point were determined to complete the grasping of unknown objects by using the principle of "grasping area and then grasping point".(5)To solve the problem of accumulated errors in the process of grasping,an attitude cognition learning system was constructed to ensure the stability of grasping,that is,the position of end-effector was adjusted in time to compensate for accumulated errors through the feedback of mechanical arm motor and depth camera,so as to ensure the smooth progress of grasping task.Experiments verify the feasibility of the grasping system in the home environment of efficient and stable grasping various items,and compared with the large amount of calculation,high cost of deep learning algorithm and the robot system using expensive hardware,it has more advantages in the home application.