一种改进机器人计算力矩控制的神经网络补偿方法

提出了一种由计算力矩控制器和神经网络补偿控制器相结合的机器人控制方案,探讨了用线性神经网络补偿机器人计算力矩不确定性误差的方法.推导了网络权值的自适应调整律,并证明了系统的稳定性和跟踪误差的收敛性.所提方案结构简单,鲁棒性强,且神经网络补偿器有较好的适应性,无需事先知道机器人动力学参数和结构的精确值.对某打磨机器人轨迹跟踪的实验结果表明所提方案具有很好的鲁棒性和抗干扰能力.

A method of an enhanced computed-torque control for robot with a neural-compensator

A novel control scheme, based on the combination of computed-torque control as a feed-forward structure and a linear neural network as a compensation structure, was proposed to improve the computed-torque control uncertainties of a polishing robot. An adaptive update law of network weights was induced, and the system stability and tracking error convergence were demonstrated. The resulting control scheme had a simple structure with enhanced robustness. Since the neuro-compensator had good adaptability, accurate knowledge of both the robot dynamic parameters and structure were not required in advance. The experimental studies on a polishing robot trajectory tracking verified the robustness and anti-disturbance capacity of the proposed control method.