一种受人类触觉感知启发的机械手指尖仿生感知系统  

Biomimetic fingertip perception system inspired by human tactile perception

作　　者：熊一萱 熊鹏文[1,2] 汪永阳 洪金华 刘小平 宋爱国[4] XIONG YiXuan;XIONG PengWen;WANG YongYang;HONG JinHua;LIU Peter X.;SONG AiGuo(School of Information Engineering,Nanchang University,Nanchang 330031,China;Jiangxi Key Laboratory of Intelligent Robot,Nanchang 330031,China;Department of Systems and Computer Engineering,Carleton University,Ottawa K1S5B6,Canada;School of Instrumentation Science and Engineering,Southeast University,Nanjing 210096,China)

机构地区：[1]南昌大学信息工程学院,南昌330031 [2]智能机器人江西省重点实验室,南昌330031 [3]Department of Systems and Computer Engineering,Carleton University,Ottawa K1S5B6,Canada [4]东南大学仪器科学与工程学院,南京210096

出　　处：《中国科学:技术科学》2025年第1期133-147,共15页Scientia Sinica(Technologica)

基　　金：国家重点研发计划“智能机器人”专项(编号:2023YFB4704900);国家自然科学基金项目(批准号:62373181,62163024,61903175,61663027);江西省“双千计划”资助项目(批准号:jxsq2023201097);江西省杰出青年基金资助项目(批准号:20232ACB212002);江西省主要学科学术与技术带头人资助项目(批准号:20204BCJ23006)资助。

摘　　要：指尖作为触觉感知最为敏感和丰富的部位,通常与交互物体直接接触,其小尺度和精细化的感知特性使其对细微触觉信息的感知和处理能力至关重要.然而传统触觉感知系统中存在多点数据处理复杂度高以及动态特征提取效果不佳等问题,为此,本文受人类触觉感知机制的启发,提出了一种机械手指尖仿生感知系统.首先,本文设计了一种可装载式仿生机械手指,模拟指尖皮肤触觉感受器的功能,采集触觉信息并构建多样本触觉数据集.其次,本文针对指尖小尺度精细化触觉感知响应机理,提出了一种基于时频定位的触觉信号耦合方法,用于处理和分析触觉数据.阵列式多触点响应信号经时空频率编码转换为复合时域信号,进一步结合短时傅里叶变换生成保留时间特性的触觉特征频谱图,模拟人类神经系统对触觉信息的整合、解耦与处理过程.最后,通过改进的卷积神经网络对二十六个英文字母进行分类实验验证.实验结果表明,本文提出的指尖仿生感知系统以及基于时频定位的触觉信号耦合方法在小尺度精细化指尖触觉时序数据处理方面具有很好的实用性,能够有效地实现不同纹理信息的物体的分类识别,分类精度达到99.61%.The fingertip is the most sensitive and richly endowed component of tactile perception,typically making direct contact with interacting objects.Its small-scale and refined perception characteristics afford it the ability to sense and process critically important subtle tactile information.However,traditional tactile perception systems suffer challenges such as high complexity in multipoint data processing and ineffective dynamic feature extraction.To address these issues,this paper proposes a biomimetic fingertip perception system for robotic applications inspired by human tactile perception mechanisms.First,a mountable biomimetic robotic fingertip is designed to simulate the functions of fingertip skin tactile receptors.It acquires tactile information and builds a diverse tactile dataset.Second,a tactile signal coupling method based on time–frequency localization is proposed.It focuses on the refined small-scale perception mechanisms of the fingertip for processing and analyzing tactile data.Spatiotemporal frequency encoding is employed to convert array-based,multitouchpoint response signals into composite time–domain ones.These signals are subjected to short-time Fourier transformation to generate tactile feature spectrograms that retain temporal characteristics.This process simulates the integration,decoupling,and processing of tactile information in the human nervous system.Finally,an improved convolutional neural network is used to classify the 26 English letters for experimental validation.The results demonstrate that the proposed biomimetic fingertip perception system and the tactile signal coupling method based on time–frequency localization are highly effective for processing small-scale,refined,tactile,sequential fingertip data.The system classifies and recognizes objects with various texture information,achieving a classification accuracy of 99.61%.

关 键 词：仿生指尖 触觉感知 仿生感知 分类识别 卷积神经网络 

分 类 号：TP2[自动化与计算机技术—检测技术与自动化装置]