Research On Autonomous Dynamic Real-time Capture Of Industrial Robot Based On Prediction Model

Vision-based robot dynamic target capture is of great significance for realizing the transformation and upgrading of the manufacturing industry and seizing the initiative in the national strategic field.This paper takes the 6-DOF industrial robot UR10 as the research object,and studies the dynamic target capture of the robot.Based on Monte Carlo method,the workspace of UR10 robot is calculated and the pose cloud images of different planes are generated.The link coordinate system of the robot is constructed,the DH parameters of the robot are obtained,and the theoretical model of its forward kinematics is established.The inverse kinematics theoretical model of the robot is established by inverse transformation method.The forward and inverse kinematics simulation models are built and verified in Matlab/Simulink.Based on the forward kinematics model of UR10 robot,six error models are established from three aspects of robot end position,attitude and pose synthesis.Through the identification simulation experiment,the error model with the highest identification accuracy and the best stability is selected for the actual identification experiment.The particle swarm optimization algorithm is used to complete the ’hand-eye calibration ’ of the UR10 robot.Aiming at the existing errors,the actual identification experiment is carried out based on the optimal error model selected by simulation,and the actual DH parameters of the robot in the camera coordinate system are obtained to ensure the positioning accuracy of the robot in the camera coordinate system.The trajectory planning of UR10 robot in Cartesian space is completed by interpolation method,and the dynamic simulation is carried out with the help of robot toolbox.In the capture experiment,it is used to plan the path between the initial point of the robot and the predicted capture point.Taking the prediction accuracy as the evaluation index,a dynamic adaptive weighting method is proposed,and a hybrid model based on the rolling prediction principle and the quadratic exponential smoothing method is constructed to predict the trajectory of the dynamic target.Aiming at the problems of obvious error accumulation effect and poor prediction accuracy in the ordinary advanced prediction method,the rolling prediction principle is introduced again,and a high-precision advanced prediction method of hybrid model is proposed to improve the accuracy of advanced prediction of dynamic target trajectory.The dynamic target capture simulation model of UR10 robot is built in Matlab/Simulink to verify the established hybrid prediction model.Kalman filter and wavelet transform algorithm are used to denoise and smooth the trajectory of dynamic target,so as to reduce the mutation in prediction.The UR10 robot dynamic target capture experiment platform is built,and the joint real-time communication of robot,host computer and optical camera is established.Combined with the capture simulation model,the system parameters such as hybrid model parameters and camera sampling frequency captured by UR10 robot are determined.The judgment correction algorithm of hybrid model is introduced to correct the prediction results of dynamic target trajectory.The dynamic target capture experiment of UR10 robot is carried out to verify the effectiveness of the hybrid prediction model in robot capture.