Research On Dynamic Object Grasping Strategy Based On Machine Vision Guidance

The robot replaces manual work,which has the characteristics of saving manpower and high efficiency,and the technology development trend of integrating robot and artificial intelligence is increasingly obvious.Stochastic dynamic target grasping fully embodies the integration of robotics and artificial intelligence technologies and contains important academic significance and engineering applications.Currently,in the process of robot grasping dynamic targets,there are problems such as inaccurate target position estimation,susceptibility to interference and poor robustness of the control strategy.This paper takes a six-axis collaborative robot as the research object,and the research content includes dynamic target grasping scheme research;target pose estimation and tracking research;vision servo control grasping research;robot dynamic target grasping experimental research,etc.The main work and research content are as follows.(1)Dynamic target grasping scheme research.Based on the characteristics of the visual grasping scenario,the design index of the grasping scheme is proposed,the selection of the camera,robot and end-effector and the selection of the target pose estimation and tracking algorithm are determined.Based on the workflow of the grasping scheme,the analysis and selection of the hand-eye structure and control methods are carried out,the advantages and disadvantages of the two grasping strategies of tracking and prediction are analyzed,and the grasping strategy of "long-range tracking and near-range prediction grasping" is proposed.(2)Target pose estimation based on improved YOLOv3 and EPnP.Based on the requirements of real-time and accurate recognition and detection of targets and estimation of target pose in the grasping process,YOLOv3 is improved by introducing CIOU position loss function and ECANet attention mechanism to improve the accuracy of target recognition and detection and provide a basis for pose estimation.The EPnP method is used to derive the mathematical model for pose estimation,establish the correspondence between the reference points of the real 3D model border and the predicted 2D projection points of the improved YOLOv3,solve the target pose,and provide a basis for the robot dynamic target grasping.(3)SiamMask-based target tracking.To address the problem that target tracking is easily interfered by background or other noise,SiamMask single-target tracking method is used to obtain the correlation between the features of the current frame and the search image through twin network model and correlation analysis,and find the position corresponding to the maximum value as the predicted position of the target to achieve the tracking of the target.Secondly,the feature map of the predicted position is substituted into the mask model to achieve the segmentation of the target and eliminate the interference of background or other noise.The experimental results show that the method has good robustness and real-time performance,which lays the foundation for robot tracking and target grasping.(4)Position-based visual servo-controlled grasping research.The camera calibration and hand-eye calibration are used to convert the position between the camera and robot coordinate systems,and the mathematical model is derived based on the position-based visual servo control error function and the PID closed-loop control method to obtain the robot end-effector motion speed.The inverse kinematics is used to solve the motion angle and angular velocity of each joint of the robot,and the RRT path planning method is used to optimise the robot motion path and realise the long-range real-time tracking of dynamic targets.The linear prediction method is used to achieve close range grasping of dynamic targets,taking into account the influence of the time delay generated by each control link on the grasping results.(5)Experimental research on dynamic target grasping by robots.A robot dynamic target grasping experimental platform was built,a vision detection,inspection system and an upper computer control system were developed,and real-time communication was established with the collaborative robot to carry out target random motion grasping experiments.The experimental results show that the proposed long-range tracking and near-range prediction grasping strategies work well.The highlights of the research work include the proposed grasping strategy of "long-range tracking and near-range predictive grasping",the innovative method of target position estimation and tracking,and the construction of a grasping control system integrating vision detection,host computer and robot real-time communication.