Robust control of multi degree of freedom robot based on disturbance observer of neural network

A multi degree of freedom (DOF) robot is a complex and variable nonlinear system, and its control performance is affected by the inherent parameters of the model itself, friction, external disturbance, and other factors. A robust control method based on neural network disturbance observer was proposed in this study to improve the effect of time-varying system parameters and external disturbances on the control system performance. A new dynamic model of robot error was constructed by analyzing the characteristics of the robot system model. The total disturbance of the system was observed and compensated online on the basis of the neural network observer, and the effectiveness of the control method was verified through simulation. Results demonstrate that the robust adaptive control method with neural network disturbance observer reduces the maximum angular displacement error by 2.7 times and the maximum angular velocity tracking error by 2.14 times compared with the control method without observer when model parameter perturbation and external disturbance are found in the system. The maximum angular velocity error is 4 and 88.6 times lower than proportional derivative (PD) compensation control and traditional sliding mode control, respectively. The neural network disturbance observer can accurately track the total disturbance of the system. The input torque of the proposed control method has a small peak torque, which is 1/8 and 1/2 times lower than the sliding mode control and PD compensation control, respectively, and the control curve of the proposed control method is relatively smooth. The proposed method provides a reference for the multi DOF robot to achieve high-precision tracking in complex and changeable environment. © 2020 School of Science, IHU. All rights reserved.