Unknown Stack Object Grasping Method Based On 3D Constructon And Reasoning

As an important capability of robots,Robot grasping technology requires intelligent progression from perception to cognition,reasoning and decision-making along with the upgrading of work tasks and dynamic changes of the scene,which makes the deep research on realizing robot autonomous grasping service become the focus of research in recent years.In the process of moving from grasping objects in simulated environment to grasping objects in real unstructured environment,it is mainly the problems of environment perception,robot cogni tion and grasping planning that hinder the development of robot grasping applications.At this stage,through the two-dimensional image visual identification combined with the depth camera positioning information to guide the robot to complete the known mod el of target capture has been widely used,but most of all is an independent or a single object,and for stack with unknown object scene grasping problem has yet pour good solved,this greatly limits the application of robots in the real world.From the perspective of environmental awareness,the research solves the problem of identifying and locating the unknown object,and add the understanding of the scene structure by reasoning space relation between the stack object,obtain the strategy of robot grab order.To avoid falling down and unnecessary damage,etc,it achieves autonomous grasping in the real world with security and stability.Firstly,the forward kinematics of the robot is used to obtain the transformation matrix of the camera,and Point cloud registration method is used to calibrate the position of the camera,improving the quality of 3d reconstruction.By integrating the deep neural network with the point cloud geometric segmentation algorithm,improving the precision of edge segmentation,the segmentation effect is further improved,the segmentation accuracy improved by about 20%,furthermore,the object semantic recognition and positioning are realized,and the fine 3d semantic segmentation map is finally obtained.Then,through geometric analysis to extract object connection points between deduce the relationship between matrix,based on the discriminant support relationships between objects on the basis of stability theorem,remove potential definite physical interactions for pseudo contact,recursive induces the object layer to build a relationship tree,grab the final output meet the target object,and through comparing the simulation verify the effectiveness of the algorithm and accuracy.The optimal grasping pose is obtained by using the grasping pose detection algorithm,and the grasping operation is realized.Finally,based on the ROS(Robot Operating System),we adopt Turtlebot3 with Open Manipulator as the operating object to conduct experiments,using Real Sense collecting scene information,and a series of algorithms proposed in this paper were simulated and verified by using Real Sense to collect information.It was proved that turtlebot3 could grasp the unknown stacked objects in a safe and stable manner and achieve intelligent progression from perception to cognition,reasoning and decision making.