Research On Robot Grasping Planning Based On RGB-D Visual Recognition

In order to detect and rescue the highly dangerous radiation environment,the intelligent robot is developed to replace the human for dangerous work,and the life safety of the people is guaranteed to a great extent.As an automatic mechanical device,the intelligent mobile robot can obtain the information of the scene and identifie the specific location of the dangerous source.This paper studies robot's autonomous grasping technology and completes the tasks of grasping,sampling and operating the radioactive objects.Aiming at the detection and recognition of the radioactive objects,a visual recognition algorithm is designed in this paper.Using the RGB-D sensor Kinect V2 to get the scene information of the radiation environment,it is transmitted to the robot control center in the form of images and point cloud.The control center is based on the computer vision library Open CV and the point cloud library PCL to complete image processing and point cloud processing.The target recognition is developed with the algorithm of ORB and template matching.As a result,the radioactive source objects are found.The point cloud registration and position estimation with LM-ICP are used to calculate the position and posture of the objects relative to the camera.Based on the position and posture information obtained by visual recognition,this paper presents a motion planning algorithm for high dimensional space manipulator,which solves the problems of high dimension,high randomness,and not optimality.Establishing of the forward kinematics models of the robot,the LBT-RRT high dimensional space motion planning algorithm is designed,written and verified under the condition of obstacles.In order to verify the visual recognition algorithm and the motion planning algorithm,ROS virtual experiment platform is built,and the simulation motion is analyzed.The RGB-D camera Kinect is used to collect the images,and the target recognition algorithm of ORB and template matching is verified.By using planning scene and collision detection,verification test of the LBT-RRT motion planning algorithm is completed.Using the sports planning library Move It! and the physical simulation environment Gazebo,the robot simulation grasping tests are successful.The comprehensive robot experimental platform is built to verify the robot's grasping ability.The hand eye calibration experiment of the robot and the camera is used to determine the relationship between the Kinect V2 and the robot.The target recognition and 6D pose estimation experiments can get the position and posture of the target object.The LBT-RRT motion planning verification tests ensure the smooth progress of grasping objects.The robot grasping operation comprehensive experiment is conducted,and robot grasps the objects of different shapes and sizes successfully.