基于多层径向基神经网络的机器人手爪腕力测量

Measurement of Robot Wrist Force Based on Multi-layer RBF Network

机器人手爪腕部所受的力的量化对于机器人的安全操作是非常重要的。为了获取手指上力传感器的输出变化,首先对机器人进行标定实验,得到了机器人在行走和操作工件时腕部所受的多维力的信息。分析了利用经典的BP网络、RBF网络方法对机器人手爪腕力测量的缺陷,设计了一种新的神经网络-3R神经网络,对机器人手爪的8个指力传感器数据进行数据融合。结果显示,该方法能大大提高手爪腕部所受的力的估测精度,从而为机器人的安全操作提供了决策依据。

Quantitative analysis of wrist force of robot grippers is very important for the safe operation of robot. In order to acquire output variety of gripper sensors of robot, calibration test must be done. Through calibration, multidimensional information of robot while it moves or operates object was acquired. Analyzing disadvantages of methods of BP and RBF neural network, a new kind of neural network-3R neural network was designed. The network structure and weight values of the 3R neural network were obtained by data fusion of outputs of finger force sensors on the robot gripper. Result proves that the accuracy of estimate of the wrist force increases consumedly by use of this method, and it can provide decisions for the safe operation of the robot.