Research On Visual Servo Control Technology Of Robotic Manipulator

The visual servo control of the manipulator is a question for study or discussion of multiple discipline cross-infiltration,which refers to kinematics,dynamics analysis,computer vision technology,machine learning theory and automatic control theory,and it is a very challenging subject.This paper uses kinematics and dynamics to be analysis tools,and uses improved extreme learning machine and deep learning as the primary technical lines,which compensates and corrects manipulator model errors and visual dynamic errors,and conducts in depth research on visual servo control technology of manipulator.Central contributions are as follows:(1)In order to solve the problem of poor real-time performance and low accuracy for inverse kinematics of manipulator based on neural network,the paper presents a solution in manipulator inverse kinematics of hybrid extreme learning machine(ELM),genetic algorithm(GA)and particle swarm optimization(PSO).Firstly,it uses the input layer weights and hidden layer bias of extreme learning machine to initialize randomly,so as to obtain the advantage of real-time performance and to obtain the initial inverse kinematics solution of the six-degree-of-freedom manipulator,and then,uses the hybrid optimization algorithm based on particle swarm optimization and genetic algorithm to optimize the initialize inverse solution,which improved the accuracy of the inverse solution.The hybrid optimization method can not only exert the advantage of strong global search ability of the genetic algorithm,but also exert the advantage of strong local search ability of the particle swarm algorithm.The experimental results show that the real-time performance of the inverse solution of the manipulator in this paper is good,and the error of the end effector in the manipulator is small.(2)In order to solve the problem of estimation difficulty and poor real-time performance of interactive matrix in visual servo control,a manipulator visual servo control system based on multivariate adaptive regression spline(MARS)and optimized extreme learning machine hybrid algorithm is proposed.It is differ to the traditional image based visual servo control system,which selects constant as the interaction matrix and its pseudo-inverse,this paper proposes a multivariate adaptive regression spline and optimized extreme learning machine hybrid algorithm(MARS-RKRIELM)for real-time estimate the pseudo inverse of the interaction matrix.First of all,the hybrid algorithm introduces a random reduction kernel and regularization parameters on the basis of the incremental extreme learning machine(I-ELM),and solves singularity problems and reduce the risk over-fitting of algorithms that the nuclear extreme learning machine(KELM)exists when the training sample is smaller than the hidden layer neuron.Then,according to the characteristics of feature importance can be selected in MARS method,and the MARS can be combined with the optimized I-ELM.Finally,through the comparison of simulation experiments,the excellent performance of the visual servo control system is verified.(3)For the hand-eye visual servo system based on Kalman filtering has the influence of disturbance error,noise statistical error and the slow convergence speed of the system,the paper proposes an online sequential extreme learning machine based on the forgetting mechanism and regularization,which has a correction to the error in the Kalman filter algorithm,at the same time,the cerebellar model algorithm is used to optimize the joint angle of the manipulator,to make the performance of the image based visual servo(IBVS)control system is greatly improved.Firstly,an online recognition model of image Jacobi matrix based on Kalman filter(KF)is established.In order to solve the error caused by the KF algorithm's disturbance to the robot system and the instability of the noise statistics caused by the colored noise,an error compensation model based on the forgetting mechanism and the regularization of the online sequential extreme learning machine(FROSELM)was established to estimate the KF.The resulting sub-optimal estimates are again optimized.Based on the joint angle obtained by image feature error feedback,an optimization method based on cerebellar model is designed.A visual servo control method(CMAC-KF-FROSELM)combining CMAC and KF-FROSELM is proposed.Compared with the existing methods,the CMAC-KF-FROSELM-based visual servo control method does not require camera parameters in the servo process,and has strong ability to resist system disturbance errors and noise statistics errors.At the same time,the optimization method of adding cerebellar models to machinery correcting the manipulator's angle significantly improves the convergence speed of the visual servo.(4)For the problem that the visual servo control system based on Kalman filter is susceptible to noise interference,the initialization of Jacobi matrix is difficult and its observation value is not accurate enough,an uncalibration visual servo control system method of combining robust Kalman filter algorithm and a long-short-term memory(LSTM)network are proposed.In the entire visual servo system,the Jacobi matrix is estimated in real time with robust KF,and the manipulator motion is controlled,at the same time,the results of RKF estimation are trained online on the LSTM network,and then,the Jacobi matrix predicted by the trained LSTM model is used for updating the robust KF state quantity.Through the comparison of system convergence speed,end effector motion trajectory length and cumulative error,the proposed IBVS control system in this paper has fewer iterations,more smooth trajectory and higher visual servo control accuracy than KF IBVS,KFANN IBVS,NNRL IBVS,and RKF IBVS.(5)Sliding mode control based on OS-Fuzzy-Dropout-ELM is proposed.For the position tracking control problem of manipulator,this paper designs a new type of intelligent control system(OFDELMISMC).The system starts from the dynamic model of manipulator,uses OFDELM algorithm to simulate the equivalent control law of sliding mode control,and updates the network parameters and fuzzy system to relax the detailed information requirements of the system through online cycle training,and uses Lyapunov stability principle to derive the adaptive learning rate of network to ensure network convergence and stable control performance.The experimental results show that the OFDELMISMC system can achieve the precise tracking of the desired trajectory of the manipulator and effectively reduce the chattering phenomenon,so it is a practical and feasible control scheme.(6)The manipulator sliding mode control of the optimized extreme learning machine based on the hybrid grey wolf algorithm is studied.Aiming at the problem that the extreme learning machine(ELM)randomly selects hidden layer weights and biases to cause its performance to be unstable,a hybrid gray wolf optimization algorithm is proposed to optimize the hidden layer weights and biases of ELM,and the improved extreme learning machine will be introduced to sliding mode control,the machine uses the approximation ability of the extreme learning machine to approximate external disturbances and uncertainties,and compensates for this.The discontinuous control signal is serialization,which effectively reduce the chattering phenomenon of the sliding mode sliding mode control.The simulation results for the two-link manipulator show that this scheme can achieve high precision,high speed tracking effect,and can suppress the existing model error and the chattering phenomenon under external disturbance conditions.