机器人臂的多项式PD型学习控制方法

Learning control method of polynomial PD-type for robot manipulator

提出一种用于机器人臂的带有重力补偿的多项式PD型(PPD)学习控制器,基于多项式神经网络给出了这种控制器的比例系数连续学习算法,由非线性机器人动力学模型与所提出的学习控制器所组成的闭环系统被证明在满足李雅普诺夫直接法和拉萨尔不变集定理时是全局渐近稳定的,除了理论结果,也提供了在两自由度机器人臂位置控制中的仿真实验比较,结果表明PPD学习控制器在系统快速响应性方面优于常规PD控制器。PPD学习控制器为机器人控制系统提供了一种新的途径。

A learning controller of polynomial PD-type （PPD） with gravity compensation for robot manipulators was presented in this paper. A continuously learning algorithm for its proportional coefficients was given based on the polynomial neural network. That the closed-loop system composed by full nonlinear robot dynamics and the proposed learning controllers is globally asymptotically stable is proved in agreement with Lyapunov＇s direct method and LaSalle＇s invariance principle. Besides the theoretical results, the comparison of simulation experiments is also presented in the position control of a robot manipulator with two degrees of freedom to illustrate the PPD learning controller is superior to conventional PD controllers in high-speed response of the control system. The PPD learning controller provides a novel approach for robot control systems.