Research On Uncalibrated Visual Servoing Control Of Manipulator Based On Reinforcement Learning

The manipulator is controlled directly by the image features of the camera image plane based on uncalibrated visual servoing control,it avoids accurate calibration of the internal and external parameters of the camera,the robustness of the manipulator control system in the actual industrial environment with interference is improved.However,the uncalibrated visual servoing controller needs to establish the nonlinear mapping between manipulator motion space and image feature space,the nonlinear mapping relationship can be estimated by using image jacobian matrix,but it's approximately estimated.Researchers have used neural networks to fit their nonlinear mapping relationship,but a lot of time is required for the selection of network models and training of data,and the model has poor anti-interference ability.For the above questions,this paper adopts the reinforcement learning algorithm derived from simulating animal learning behavior,no sample data is required for training,it allows the manipulator to study online and in real time in the actual environment,and keep the manipulator closer to the optimal motion path while working.The main research work is as follows:(1)The uncalibrated visual servoing control of manipulator based on reinforcement learning is proposed.The visual servoing controller is mainly composed of reinforcement learning algorithm and Kalman filtering algorithm,first of all,kalman filtering algorithm is used to estimate the image jacobian matrix,when the image feature space and the manipulator joint space are slightly changed,the nonlinear mapping relationship between them is established in the form of approximate linearization,and then the temporal difference reinforcement learning algorithm is used to train the visual servoing controller of manipulator online and in real time,the visual servoing controller is iteratively optimized,and the visual servoing controller approach the optimal control strategy by iterative learning,to make the manipulator complete visual servoing control task.(2)The Robot Toolbox and the Machine Vision Toolbox is used to build the simulation environment for the uncalibrated visual servoing control of manipulator based on temporal difference reinforcement learning algorithm in MATLAB simulation platform,and the simulation experiment is carried out in the simulation environment,the validity of the uncalibrated visual servoing control based on reinforcement learning algorithm is verified.(3)The physical verification platform based on binocular fixed visual layout is set up.IPC?small six-DOF industrial manipulator and RC7 M controller produced by Denso robot company is used to establish the manipulator control system,the binocular fixed visual layout system is built by Daheng CCD industrial camera,in addition,the physical experiment of uncalibrated visual servoing control of manipulator based on temporal difference reinforcement learning algorithm is completed on physical platform,and the validity of the proposed algorithm is verified.