Dynamic Learning From Adaptive Neural Control of Robot Manipulators With Prescribed Performance
针对未知动力学和外部干扰的n连杆机器人操作器，提出一种自适应神经控制方案，通过误差变换和径向基神经网络实现规定跟踪误差性能，并利用经验知识实现静态神经学习控制，避免在线参数调整。
This paper presents dynamic learning from adaptive neural control (ANC) with prescribed tracking error performance for an n-link robot manipulator subjected to unknown system dynamics and external disturbances. To achieve the prescribed performance, a performance function is introduced to describe the performance restrictions on tracking errors, and then specific performance requirements are served as a priori condition of tracking control design. By an error transformation method, the constrained tracking control problem of the original robot manipulator is transformed into the stabilization problem of an unconstrained augmented system. Then, a novel ANC scheme is proposed for the unconstrained system by combining a filter tracking error with radial basis function (RBF) neural network (NN) approximator, and all the signals in the closed-loop system are semi-globally uniformly ultimately bounded. The external disturbances might make it difficult to achieve the accurate convergence of NN weight estimates. To overcome this difficulty, an appropriate state transformation is introduced to transform the closed-loop system into a linear time-varying system with small perturbed terms. Under partial persistent excitation condition of RBF NNs, the convergence of NN weight estimates is guaranteed, and then the experienced knowledge on the unknown robot manipulator dynamics can be stored with NN constant weights. Using the experienced knowledge, a static neural learning control is proposed to improve the system performances without time-consuming online parameter adjustment process, and the proposed learning control can also guarantee the prescribed transient and steady-state tracking control performance. Simulation results demonstrate the effectiveness of the proposed method.