Research On Robot Dexterous Hand Grasping Technology Assisted By 3D Vision

In today's China,where the labor force is sharply declining and aging is intensifying,it is the best choice to get rid of development difficulties with automation technology.Current automation technology has played a major role in the national economy,but there are also problems of insufficient flexibility.Taking the “pick up and place” as an example,the traditional method requires the operator to manually teach the robot move point by point,which is only applicable to the structured scene where the scene and the target object are unchanged.Modern production is usually “small batch,multi-process,multi-category”,witch the traditional method cannot meet the demand.Based on the new requirements of modern flexible manufacturing for robot automation,this paper focuses on machine vision and studies the related technology of visual aided grasping.It mainly includes:1)Three-dimensional modeling of objects: In order to obtain the object's model conveniently and cheaply,a simple method for the object's three-dimensional modeling is realized based on the rotating platform and RGBD camera.By fixing the RGBD camera and collecting 15-25 images of the object on the rotating platform,a threedimensional model for visual aided grasping can be constructed.2)Hand-eye calibration: The dual quaternion hand-eye calibration method is deeply studied,and a one-button automatic hand-eye calibration method is designed to reduce the requirements for operators.Experiments show that the method can quickly and better complete hand-eye calibration.3)Pose estimation: Combined with deep learning technology,a pose estimation algorithm without manual labeling is proposed.Firstly,the image is segmented by Mask-RCNN,which has been proved to be successful in segmentation,so as to reduce the search space.Then,for the segmented point cloud,hypothesis generation,hypothesis verification,and refinement processing are performed to obtain the pose estimation of the target object.4)Visual aided grasping system: With UR5 manipulator,Robotiq grasp,ASUS RGBD sensor as hardware platform,ROS and MoveIt! as software platform,a visual aided grasp system is constructed.A sampling-based grasp planning method is implemented for the parallel two-finger gripper used in this system.The three-dimensional modeling of objects,hand-eye calibration,pose estimation and the construction of visual aided grabbing system studied in this paper basically cover the main technical problems involved in robotic automated grabbing tasks.At the same time,the improvements to overcome the current shortcomings of these technical problems can improve and accuracy and the ease of use in practice.