Research On Tracking Control Of Robot Dynamic Target Grabbing

The robot's grasping of moving objects is an innovative and challenging task.It requires the robot to detect the moving object autonomously,obtain the current position of the object and accurately predict the trajectory of the moving object.On this basis,the robot's own motion is planned and controlled to achieve successful grasp of the moving object.An overall scheme is proposed for the operation of a seven-degree-of-freedom robot to grasp moving object.The main research contents include the following four aspects:(1)Modeling of the Seven-degree-of-freedom robot.The research object of this paper is a Sawyer seven-degree-of-freedom robot of Rethink.According to the D-H method,the reference coordinate system of each joint of the robot is defined,and the corresponding D-H parameters are calculated.The kinematics and dynamic equations of the robot are derived,and the correctness of the modeling is verified using the Robotics System Toolbox in Matlab.(2)Moving-object detection based on convolutional neural network.The moving image of the object is obtained through a binocular vision system,and the structure of the convolutional neural network is designed and trained.The pre-trained model is used to detect the center position of the object's grasping rectangle and predict the success probability of grasping from different directions,so that the suitable grasping rectangle for the robot is found.The FAST feature points is extracted in the region of the detected grasping rectangle,and then the matching between the left and right camera images is obtained.(3)Vision-based object localization and trajectory prediction.According to the coordinates of the feature points obtained by image matching in the image,the visual positioning of the object is completed by combining the method of camera space manipulation(CSM).In this way,the three-dimensional spatial position of the moving object in the robot coordinate system is obtained.CSM method can ensure good positioning accuracy without pre-calibration of camera parameters,while in traditional visual object positioning methods,it is necessary to accurately calibrate the camera in advance.The trajectory data of the object motion is used to train the support vector regression(SVR)model to characterize the motion characteristics of the object.The obtained three-dimensional spatial position of the moving object in the robot coordinate system is taken as the initial state,and the support vector regression model is used to predict the object's moving trajectory.(4)Robot trajectory planning and tracking control.The motion of the robot is planned in combination with the predicted object motion trajectory.Combining the kinematics model of the seven-degree-of-freedom robot,the desired robot joint angle is obtained by inverse kinematics solving the predicted target position.Considering the kinematics and dynamics constraints of the robot,a second-order continuous trajectory curve in the robot joint space is planned to ensure the smoothness of the robot motion during the grasping process.Then the compute torque controller is designed based on the dynamic model of the robot to complete the trajectory tracking of the robot.The simulation is carried out in Matlab.The simulation results show that the designed controller can ensure that the robot moves accurately along the planned trajectory and successfully completes the grasping operation of moving object.