基于注意力双向门控循环神经网络的索驱末端效应器运动估计  

Motion estimation for cable-driven distal end-effectors using attention-based bi-directional gated recurrent neural networks

作　　者：徐新洲 陈永发 刘光明 李紫千 赵力[3] 王正雨 Xu Xinzhou;Chen Yongfa;Liu Guangming;Li Ziqian;Zhao Li;Wang Zhengyu(School of Internet of Things,Nanjing University of Posts and Telecommunications,Nanjing 210003,China;School of Mechanical Engineering,Hefei University of Technology,Hefei 230009,China;School of Information Science and Engineering,Southeast University,Nanjing 210096,China)

机构地区：[1]南京邮电大学物联网学院,南京210003 [2]合肥工业大学机械工程学院,合肥230009 [3]东南大学信息科学与工程学院,南京210096

出　　处：《Journal of Southeast University(English Edition)》2023年第2期187-193,共7页东南大学学报（英文版）

基　　金：The China Postdoctoral Science Foundation(No.2022M711693);the Natural Science Foundation of China(No.52175221,61801241,62071242);the Natural Science Foundation of Jiangsu Province(No.BK20191381).

摘　　要：为实现对索驱末端效应器运动状态的自适应估计,基于注意力双向门控循环神经网络,提出了一种数据驱动的末端效应器运动估计方法.首先进行数据构建,以获取短时序列作为训练样本.然后,将数据输入包含带自注意力模块的双向门控循环神经网络,构造样本的序列模型.最后,在索驱末端效应器运动数据集上,将电机位置、速度以及作为系统控制信号的输入时间序列作为样本特征,进行运动估计性能对比实验.结果表明,相比常用的序列建模回归算法,所提方法能够取得更好的末端效应器运动估计性能,因而能有效实现复杂条件下对索驱末端效应器的运动估计.A data-driven motion estimation approach based on attention-based bi-directional gated recurrent neural networks was proposed to adaptively estimate the motion of a cable-driven distal end-effector.First,the data construction was performed to obtain short-term temporal sequences as training samples.The data were then processed using the bi-directional gated recurrent neural networks with self-attention modules for building sequential models on the samples.Finally,based on the motion dataset of the cable-driven distal end-effectors,the estimation-performance comparison experiments were performed using the motor s position,speed,and input time sequence for the system-control signal as the sample features.The results show that compared with conventional sequence modeling regression approaches,the proposed approach can achieve better performance for estimating the motion of the end-effector.Therefore,it can effectively estimate the motion of cable-driven distal end-effectors under complex conditions.

关 键 词：索驱末端效应器 运动估计 双向门控循环神经网络 注意力机制 

分 类 号：TP241.3[自动化与计算机技术—检测技术与自动化装置]