Research On Robot Intelligent Grabbing Syetem Based On ROS Visual Positioning

With the development of science and technology,robots are used more and more widely.Nowadays,the control of robots in industrial production line is mostly on-line teaching and off-line programming.However,in view of the problem that the initial and final postures of grasping target objects are strictly limited,robots can only achieve point-to-point operation mechanically.The whole process takes a long time,the operation efficiency is low,and the positioning accuracy is low.Therefore,the robot is somewhat lacking in intelligence.Based on this problem,this paper proposes a robot intelligent grabbing system based on ROS visual positioning to solve the problem that the initial pose and final pose of the grab target are strictly limited.Firstly,obtaining its internal and external parameters,Zhang Zhengyou algorithm is used to calibrate RGB-D camera.Secondly,checkerboard lattice and AR tag are adopted to achieve hand-eye calibration.Then,object recognition and pose estimation are carried out through multi-mode information.Finally,the pose of the object is obtained by visual positioning,and the pose of the object is transformed into the world coordinate system through hand-eye calibration data,so as to realize robot grasping.The specific research contents of this paper are as follows:(1)The robot intelligent grab system based on ROS vision positioning is built,the hardware architecture and control process are designed and the tasks of controlling hardware and processing image data are completed by upper computer programming.(2)Aiming at the problems such as single source of image information,complex processing process and positioning error in current object recognition algorithms,a method of vision recognition and location based on multi-modal information is proposed.RGB-D camera is calibrated on ROS operating system to obtain its internal and external parameters in order to correct camera alignment and ensure that points in the space scene are fully projected onto the image;The recognition and location of cola bottle is realized by using the visual positioning algorithm based on multi-modal information.(3)In order to solve the problem of blind acquisition path and computation speed in the basic fast random search tree(RRT)algorithm,an improved RRT algorithm is proposed,which can increase the success rate of large-scale and high-dimensional motion planning space.After the motion planning is completed,the position relationship between the RGB-D camera and the robot base is calibrated by two methods of acquiring checkerboard pose and AR marking pose by RGB-D camera.At last,Through the comparison and analysis of rviz visual interface and real scene in ROS system,it is determined which hand-eye calibration results are more consistent with the real scene in the laboratory.The experimental platform and simulation environment of the robot intelligent grab system are built.The results of hand-eye calibration are obtained by two methods.Through the comparison and analysis of rviz visual interface and real scene in ROS system,it isdetermined which hand-eye calibration results are more consistent with the real scene in the laboratory.Then,the robot arm is visually captured,which the experimental results show the operability and feasibility of the robot intelligent grasping system.