Research On Robot Visual Servo Based On Nerual Network And Sliding Mode Control

The research of manipulator control model has entered a hot era in recent years.Compared with the past and present system models,the robustness,accuracy,and flexibility have been greatly improved and improved.At present,to solve the nonlinear problem of the visual servo of the manipulator,it is proposed to combine the Jacobian matrix,the image feature error,and the motion speed of the manipulator to complete a nonlinear mapping of the image characteristics of the manipulator and the joint angle signal.However,the accurate calculation of the image Jacobian matrix and the adaptation of the speed controller face a difficulty.In addition,the sliding mode control of manipulator combined with different algorithms has solved many problems encountered in manipulator modeling,such as the approximation of the unknown sliding mode term by radial neural network,and the approximation of the switching surface in the sliding mode controller by fuzzy method.However,there are still problems with poor network generalization ability,slow training time,and the influence of fuzzy algorithm membership function selection on system stability.In response to the problems in the current research,this article proposes the following two improved methods after reading a large number of Chinese and foreign documents:(1)For the model of uncertain camera calibration and 3D space geometry,the state space is added to the model of the depth-independent Jacobian matrix to establish the "vision" to "motion" space to complete the online estimation of uncalibrated visual servoing.In the on-line estimation process of the manipulator,the image feature error caused by the interference of external noise and camera parameters affects the solution of the Jacobian matrix,resulting in the non-linear mapping between the manipulator joint angle and the image feature.This article combines the Kalman filter to establish an error estimation model,and proposes an improved neural network and adaptive gain method to solve this problem.First,the RVFL neural network is regularized,and then Dropout is added to complete the random shutdown of the network nodes,thereby improving the generalization and adaptability of the neural network.The generalization and adaptability of the neural network are conducive to the robustness of the IBVS system.In addition,in order to solve the convergence problem of the end effector of the manipulator,the adaptive gain is proposed in this paper.The velocity and characteristic error of the end trajectory of the manipulator are used to complete the adaptive gain,so as to ensure that the chattering amplitude of the end effector of the manipulator is reduced and the convergence speed of the system is accelerated.Finally,the simulation results show that the IBVS with this algorithm has faster convergence speed and better robustness.(2)In order to solve the instability problem of manipulator sliding mode control system due to the external uncertain friction,disturbance and load change,this paper designs a new sliding mode controller which combines adaptive rules,fuzzy control method and Random vector functional link network(RVFL).The RVFL is fuzzified to improve its generalization ability.In the process of fuzzification,the fuzzy rules and hidden layers are self-mapping,which reduces the complexity of the algorithm.Then,the self-adjustment of the output weight is carried out by the adaptive rule,so as to realize the accurate approximation of the dynamic unknown nonlinearity in the control system.In addition,a robust control item is added to the controller to ensure that the chattering of the sliding mode control is in a smaller range and has a better fit for the movement of the two joints.Finally,the Lyapunov theorem is verified in the controller to ensure the closed-loop stability of the system and under the Window10 system environment,MATLABR2015 a is used to simulate and compare its models to show the effectiveness of the algorithm.