Research On Recognition,location And Control Method Of Visual Servo Mobile Manipulator Grasping

Visual servo control has attracted considerable attention from many scholars and technicians due to its substantial information,robustness,and high adaptability.However,existing research on visual servo-based control systems of mobile robots mostly regard the control of the arm and the mobile platform as two independent subsystems.As such,the results are limited.Moreover,the low location accuracy leads to manipulator grasp failure.When visual calibration is performed in the two subsystems separately,the grasping point of target pose estimation and the stable convergence of manipulator motion control are eventually affected by accumulation errors during the robot's continuous motion.Such errors are due to inaccuracies in the mechanical processing and assembly of the robot system model.Therefore,this study explores the target identification,location and control process in the above aspects through the following methods:1)Study on the self-calibration method of the eye-in-hand and eye-to-hand binocular visual servo systems.Position-based visual servo(PBVS)arm approach motion and image-based visual servo hand grasp alignment motion are proposed for the robot to grasp convex objects.This study presents a dual vision structure of fixed and local vision that cannot be adjusted on the baseline of a joint mobile platform.2)Research on the design method of moveable manipulator visual tracking control based on image moment feature points.After the analysis of the mobile robot's visual servo system control method,a visual servo motor control method for a six-degree-of-freedom robot is presented.The method using zero and the first moment of the x–y axis are used to determine the center of the target plane.The linear and angular velocities of eye-in-hand vision express the changes in the extracted feature points.In consideration of the relationship between the image feature vector increment and joint angular velocity variable,the manipulative motion controller is designed using the image Jacobian matrix.The trajectory of moment feature points is captured experimentally.Compared with the trajectory of visual servo control using other image features,that of the proposed controller is more advantageous and effective.3)Research on the robot's active flexible motion control method based on adaptive stress/position hybrid theory.When a robot completes the coarse positioning movement,the trajectory tracking control based on PBVS.Research on force feedback information is conducted to achieve the control method of compliance.Such method can reduce the accuracy requirement on a target's grasping pose through a closed-loop control structure.In accordance with the problem of multisegment trajectory between the straight line and arc of the manipulator's end effector,a kind of transition trajectory obtained through the five-order Bessel curve method is proposed.In addition,a force/position hybrid control method combined with a neural network adaptive algorithm is proposed under unknown geometric constraints.The method is a double closed-loop control strategy of force and position at the end effector.An experiment is conducted to program the transition expectation trajectory using an ROS RViz plugin.Then,it takes the simulation in virtual environment by Move It assistant in the design of the movement of virtual.The export data of joints,which are obtained from the Rqt_plot,reveal that the improved results of the experiment is consistent with the results in the MATLAB simulation.Compared with five times more considerable interpolation method,the rate peaks of joint angular velocity is smoothed,and the curve gradient is optimized.Lastly,the experimental results show that joint torque dynamic response accelerates,similar to the steps from NN network learning.4).Research on the target recognition and detection methods based on deep learning.For high-temperature target recognition and detection methods,a supervised learning method based on a faster RCNN network structure is proposed and used to judge the isothermal region of a thermal infrared image.For obtaining thermal imaging data in infrared array from near-infrared vision,the image database is used to detect the cover of a magnesium-deoxidized pot,humans,and other classified targets.Lastly,in the reduction reaction of industrial magnesium,the designed robot is used to recognize and locate the high-temperature target from the pot.In addition,combin with the LSTM network to solve the problem of object recognition in images with low signal-to-noise ratio(SNR).A robot is used to remove the high-temperature target from the tank and solve the problem of target identification in low-SNR images.To sum up,the theoretical and relevant experimental results show that the method can increase the working space by adding manipulator in the process of grasping tasks of mobile robot,and also can ensure it's the positioning and control accuracy.Proposed method is aimed at the recognition,positioning and control of convex object grasping target.Finally,the grasping position determined by the target's estimated pose is shown in the error analysis of X and Y axes respectively,which indicates that the error of recognition and positioning accuracy is controlled within 10 mm,so that the success rate of grasping action achieved by the end two-degree-of-freedom gripper is improved.