基于改进GAT的协作机器人装配意图识别研究  

Research on assembly intent recognition of collaborative robots based on improved GAT

作　　者：贾程翔 苗鸿宾[1,2] 薛锦辉 张灿 Jia Chengxiang;Miao Hongbin;Xue Jinhui;Zhang Can(School of Mechanical Engineering,North University of China,Shanxi Taiyuan,030051,China;Shanxi Province Deep Hole Machining Center,Shanxi Taiyuan,030051,China)

机构地区：[1]中北大学机械工程学院,山西太原030051 [2]山西省深孔加工工程技术研究中心,山西太原030051

出　　处：《机械设计与制造工程》2025年第4期87-92,共6页Machine Design and Manufacturing Engineering

基　　金：中北大学研究生科技立项(20231912)。

摘　　要：针对人-机交互模型非欧几何特征显著、传统机器人意图识别方法不能完全适配人机协作系统,导致识别准确度不高的问题,对人机协作中的模态输入形式进行扩张,并采用傅里叶KAN对传统图注意力网络进行改进,以提高意图识别精度。改进模态输入后GAT-KAN的准确率提高了4.2%,相较于改进前的方法具有更优良的预测性能。通过AUBO-i5机械臂、电动夹爪、相机以及六轴力传感器等设备搭建了一套人机协作系统,实验结果表明,该系统能够准确识别操作人员的意图并协助装配,具有良好的应用前景。The traditional robot intention recognition methods are not fully adaptable to human-machine collaboration systems due to the significance of non-European features in human-machine interaction models,which results in low recognition accuracy.In this paper,the modal input in human-machine collaboration is expanded,and Fourier KAN is utilized to enhance the traditional graph attention network,thereby improving the accuracy of intention recognition.The GAT-KAN's accuracy increases by 4.2%after the modal input is improved with exhibiting better prediction performance.A man-machine collaboration system is constructed for an AUBO-i5 manipulator,electric gripper,camera,and six-axis force sensor,among other components.The experimental results demonstrate that the designed system can accurately identify the operator's intention,assist in assembly,and hold a promising application prospect.

关 键 词：装配 人机协作 意图理解 神经网络 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]