Active disturbance rejection control with multilayer perceptron compensating network for robot systems

In this paper, a joint controller based on active disturbance rejection control (ADRC) method is designed for robot manipulators. The proposed method can overcome the deficiencies that exist in traditional methods such as weak anti-disturbance ability, control performance limited by the modeling accuracy, dynamic performance and steady-state performance difficult to balance, and complex control law design. A complete controller design method and parameter tuning method are given for the physical robot system. Besides, different optimization functions according to the control objective are designed to optimize the control parameters. A deep multilayer perceptron (MLP) is designed to compensate the modeling uncertainties based on system parameter identification. Both numerical simulation and experiment indicate the validity of the proposed method, the results show that the ADRC controller with MLP compensating network can realize the robot rapid and stable trajectory tracking, and have excellent control precision and strong anti-disturbance ability. Meanwhile, as not depending on accurate system model, that reduces the difficulty of the practical design and application, makes the proposed method having great value in engineering application. © 2020, Editorial Department of Control Theory & Applications South China University of Technology. All right reserved.