Design Of Uncalibrated Visual Picking Control System For Omnidirectional Mobile Manipulators

In recent years,robot technology has developed by leaps and bounds.Mobile manipulator is a compound robot composed of mobile robot and multi joint manipulator.Guided by visual sensor,it can replace human beings in material handling,parts assembly,storage and transportation and other tasks.It plays a huge role in the future manufacturing and logistics industry.The research focus of this thesis is to design a vision picking system of omnidirectional mobile manipulator by using visual servo theory and deep learning method.Through image-based uncalibrated visual servo,the mobile manipulator has better flexibility and adaptability in the face of various working conditions and interference.This thesis mainly includes four parts: kinematics and vision system modeling of omni-directional mobile manipulator,target attitude estimation and image feature acquisition,design of uncalibrated visual servo control system and construction of experimental platform.Firstly,the robot Jacobian matrix and the image Jacobian matrix are obtained respectively through the kinematic model of omni-directional mobile manipulator and the camera projection model,and then the composite Jacobian matrix is obtained.The matrix establishes the mapping relationship between the velocity of each actuator and the change of image features,which is the basis of visual servo control law design.Secondly,the desired image features of the visual servo system are obtained through the pose estimation of the picked object and the end effector.In this paper,dope,a depth target pose estimation method based on full convolution depth neural network,is used to synthesize the virtual data set training network model in the virtual engine according to the three-dimensional model of the target to realize the pose estimation of the target,and the classical color segmentation method is used to calculate the pose of the end effector.Thirdly,a mobile manipulator uncalibrated visual servo control system based on composite Jacobian matrix is designed by using the hybrid vision scheme of "eye in hand+ eye in chasis".The two uncalibrated visual servo controllers of the mobile platform and the manipulator cooperate to make the remote object enter the workspace of the manipulator,and then carry out the visual picking operation.Finally,the research and development of omni-directional mobile manipulator experimental platform is completed.The control system software is based on ROS framework,and the visual servo control law is realized by MATLAB.The algorithm obtains the virtual image of gazebo through the ROS interface,controls the simulation robot to carry out the simulation experiment,and then replaces the interface with the ROS drive of chassis and arm to carry out the physical experiment.Simulation and physical experiments verify the effectiveness of the method and achieve the desired control objectives.