An active disturbance rejection controller for a 2-DOF underwater manipulator is proposed in this study. The manipulator system has the characteristics of non-linearity, strong coupling and uncertainties within its model. Its operation is often disturbed by unknown water currents, so the anti-disturbance control of underwater manipulators has always been a challenge. The proposed controller in this study basically does not rely on the precise mathematical model of the object, and the model even can be decoupled. This method can eliminate the influence of model errors, time-varying parameters and external interference on the control effect. First, the entire manipulator treats different joints as several subsystems. For each joint subsystem, hydrodynamic forces, coupling terms between joints and unknown environmental disturbances are regarded as total disturbances. Subsequently, an extended state observer was designed to estimate and compensate for the total interference. In order to improve the disturbance observation effect of the extended state observer, the inertia matrix of the system is used to decouple the static part. Finally, the effectiveness of the proposed method is verified by both simulation and experiments. From the comparisons, it is confirmed that the quality of our controller in the presence of a certain inertial matrix error is better than traditional PD and continuous sliding mode control in terms of accuracy, dynamic characteristics as well as robustness.

本研究提出了一种用于二自由度水下机械手的自抗扰控制器。机械臂系统的模型具有非线性、强耦合、不确定性等特点。其运行经常受到未知水流的干扰，因此水下机械臂的抗干扰控制一直是一个挑战。本研究中提出的控制器基本上不依赖于对象的精确数学模型，并且该模型甚至可以解耦。该方法可以消除模型误差、时变参数和外界干扰对控制效果的影响。首先，整个机械臂将不同的关节视为几个子系统。对于每个关节子系统，水动力，关节和未知环境扰动之间的耦合项被视为总扰动。随后，设计了扩展状态观测器来估计和补偿总干扰。为了提高扩展态观测器的扰动观测效果，利用系统的惯性矩阵对静态部分进行解耦。最后，通过仿真和实验验证了所提方法的有效性。通过比较证实，在存在一定惯性矩阵误差的情况下，我们的控制器在精度、动态特性以及鲁棒性方面优于传统PD和连续滑模控制。为了提高扩展态观测器的扰动观测效果，利用系统的惯性矩阵对静态部分进行解耦。最后，通过仿真和实验验证了所提方法的有效性。通过比较证实，在存在一定惯性矩阵误差的情况下，我们的控制器在精度、动态特性以及鲁棒性方面优于传统PD和连续滑模控制。为了提高扩展状态观测器的扰动观测效果，利用系统的惯性矩阵对静态部分进行解耦。最后，通过仿真和实验验证了所提方法的有效性。通过比较证实，在存在一定惯性矩阵误差的情况下，我们的控制器在精度、动态特性以及鲁棒性方面优于传统PD和连续滑模控制。