Research of Visual Servoing in 4-DOF Robotic Manipulator
|School||Harbin Institute of Technology|
|Course||Control Science and Engineering|
|Keywords||robotic manipulator visual servo motor drive neural network|
On the automatic production process in modern industry, following the enhancement of computer and electronic device, robots application is more and more extensive, the tasks become more and more complex, and requirements of robot become more and more high. In order to observe the environment, most modern robots are equipped with different kinds of sensors, such as cameras(or more generally, visual sensors), sonar sensors, and tactile sensors etc. Although certain applications may require the use of particular sensors, we note that vision-based sensors provide a very powerful and versatile mechanism for sensing in general-purpose applications. A plausible research direction, therefore, is to utilize visual feedback loop control method in robot control.Visual servo of robot which is a complex non-linear control, is a complicated domain that is about image processing, computer vision, kinematics, dynamics, control theory and real-time calculation. This thesis designs a 4-DOF visual servoing robotic manipulator system of eye-in-hand model. The aim of experiment is to insert the axis to the hole. This thesis presents an image-based visual servoing control method. The method skips the complex calculation of transferring image features to robot pose, and hence gets the joint control signal directly from the information of the image plane.Robotic manipulator system is composed by joint servoing system and visual servoing system in this thesis. Joint servoing system is realized in host control computer while visual servoing system in image processing computer. A circuit board is designed to drive and control DC servo motor. The circuit board contains PWM variable-speed, limitation protecting, signal detecting and processing circuit etc. A real-time simulation system based RTW toolbox is designed by using PCI1711 card, and joint model is built. Dual-loop control of velocity and position is adopted as servo controller of joint. Image acquisition, image pre-processing and feature extracting are realized by image processing toolbox in Matlab. Visual servo controller is designed by neural learning method. This method skips the complex calculating and singularity problem in solving image Jacobian matrix. The non-linear relation between image feature error and joint angle error is learned by two neural network of far and near distance. Samples are collected and networks are off time trained in the thesis. Serial port communication program is produced to realize the intergrated test between host control computer and image processing computer, and aim of experiment is achieved. Accuracy of intergrated test is given by the terminate errors between image real features and expecting ones.