Design Of Manipulator Based On Model Learning And Linear Quadratic Optimal Control

The multi-degree-of-freedom robotic arm is one of the widely used robots and has many advantages such as high accuracy and flexibility,and is used to realize the requirements of automated industrial production assembly,medical service,military and space fields.Due to changes in the center of mass and friction introduced by the actual construction process of the robot arm,the derived model established by traditional dynamics will have large deviations from the actual model.How to use the known data to establish the joint model to avoid dynamic modeling errors is worth studying.The problem.From the aspect of controller design,the classical controller design method greatly depends on the designer's experience in its parameter tuning.Because the multi-degree-of-freedom robotic arm has characteristics such as multiple inputs,multiple outputs,and extreme nonlinearity,the classical controller design method is difficult.Satisfy the requirements of intelligent and efficient robot control.This paper proposes to design joint manipulator controller and whole arm control algorithm based on model learning and linear quadratic optimal control method.The purpose is to reduce model deviation through model learning,simplify controller design difficulty and improve control accuracy.Firstly,in the design of the robotic arm controller,the use of AR autoregressive model to establish the model regression matrix equation in the joint model of the robot arm is used to solve the joint model parameters.Subsequently,a joint quadratic LQR robotic joint controller based on an augmented matrix was constructed.The value iteration method was used to optimize the quadratic performance index of the system,and the robot arm joint model was optimally controlled.Experiments show that the controller designed by this method is in the actual control process,the robot arm system responds quickly,the steady-state time is short,and the control performance and precision are good.Then,for the problem that the complexity of the internal structure and the asymmetry of the geometric structure are difficult to measure accurately in the actual manipulator system,the gravity compensation model is analyzed from the arm structure of the seven-degree-of-freedom manipulator.Using the model learning method,a gravity compensation method based on fast calibration of data is designed to solve the gravity compensation under the condition that the center of gravity of the manipulator is unknown relative to the position of the arm.At the same time,the simulation and verification of the gravity compensation algorithm and the optimal solution are taken.Later,another idea was put forward for the model-based gravity compensation algorithm above.Under the condition that the robot arm's model parameters and dynamic modeling are not known,the Jacobian matrix is not known and a completely data-based optimization solution is used to solve the regression model parameters.The manipulator gravity compensation algorithm.This method uses the L1-norm regularization convex optimization method to screen out the base of the manipulator gravity compensation equation as the gravity compensation coefficient,which simplifies the work complexity and is not affected by the manipulator model parameters.The training error and test error were calculated using modeling data and test data,and the validity and accuracy of the method were verified.Finally,based on the ROS robot secondary operating system,the joint configuration,limit position and appearance information of the seven-degree-of-freedom robotic arm are described,and the simulation robot arm simulation environment is established.Using Linux-based and Xenomai real-time operating systems and conducted core performance comparison tests.Then the controllers of each joint dynamic model were designed and joint position control experiments were performed.By completing the whole arm control experiment,the correctness of the previous controller design and gravity compensation algorithm was verified.The results show that the manipulator control system can adjust the joint pose and follow the target signal well,which proves the effectiveness of the gravity compensation method and augmented linear quadratic control in the manipulator system.