Neuromorphic approach to tactile edge orientation estimation using spatiotemporal similarity

被引:7
作者
Kumar, Deepesh [1 ]
Ghosh, Rohan [1 ]
Nakagawa-Silva, Andrei [2 ]
Soares, Alcimar B. [2 ]
Thakor, Nitish, V [1 ,3 ]
机构
[1] Natl Univ Singapore, Inst Hlth N1, Singapore 117456, Singapore
[2] Univ Fed Uberlandia, Fac Elect Engn, BR-38400902 Uberlandia, MG, Brazil
[3] Johns Hopkins Univ, Dept Biomed Engn, Baltimore, MD 21218 USA
关键词
Tactile sensing; Neuromorphic; Biomimetic; Spatiotemporal; Edge orientation estimation; Mechanoreceptors; HUMAN HAND; FINGERTIP; MODEL;
D O I
10.1016/j.neucom.2020.04.131
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Moving a finger around the boundary edge is one of the important strategies followed by humans for ascertaining the shape and size of an object by touch. Tactile response from thousands of mechanorecep-tors in the human hand offers a high spatiotemporal resolution to perceive the edge orientation quickly (< 50 ms) and accurately (acuity of < 3 degrees). Inspired by the computational efficiency of biological tactile system, we present a neuromorphic approach to artificial tactile sensing that mimics the spike-based spa-tiotemporal tactile response of Fast Adapting type I (FA-I) mechanoreceptors. We propose a novel, model -based spatiotemporal correlation matching method to estimate the orientation of the boundary edge while a piezoresistive tactile sensor array attached to robotic arm palpates over the object. Results high-light the ability of the proposed method to efficiently leverage spatial and temporal information, by obtaining very precise orientation estimates (+/- 1.67 degrees error for edges oriented from 10 degrees to 90 degrees, with a step of 5 degrees) in spite of a low-resolution sensor (169 mm(2), 4 x 4 resolution). A comparison with both spatial and spatiotemporal based classifiers indicates that the proposed method achieves 20% lower mean absolute error (MAE) than its closest counterparts, all of which required supervised training. Furthermore, we show that even with a ten-fold loss of spike time precision (1 ms-10 ms), MAE is maintained at 2 degrees. This work highlights that even with a modest sensor size and resolution, spatiotemporal similarity met-rics can be used to obtain very precise estimates of orientation. Such an approach has potential applica-tions towards improving the tactile sensing capability in robotic/prosthetic hands where knowledge of spatial edge orientation information is paramount for object manipulation, and sensor contact area is often sparse and small. (c) 2020 Elsevier B.V. All rights reserved.
引用
收藏
页码:246 / 258
页数:13
相关论文
共 41 条
[1]   USING TACTILE DATA FOR REAL-TIME FEEDBACK [J].
BERGER, AD ;
KHOSLA, PK .
INTERNATIONAL JOURNAL OF ROBOTICS RESEARCH, 1991, 10 (02) :88-102
[2]  
CHEN N, 1995, IROS '95 - 1995 IEEE/RSJ INTERNATIONAL CONFERENCE ON INTELLIGENT ROBOTS AND SYSTEMS: HUMAN ROBOT INTERACTION AND COOPERATIVE ROBOTS, PROCEEDINGS, VOL 3, P386, DOI 10.1109/IROS.1995.525914
[3]   Active vision in robotic systems: A survey of recent developments [J].
Chen, Shengyong ;
Li, Youfu ;
Kwok, Ngai Ming .
INTERNATIONAL JOURNAL OF ROBOTICS RESEARCH, 2011, 30 (11) :1343-1377
[4]  
Dahiya R.S., 2013, TACTILE SENSING TECH, P79
[5]  
Delhaye B.P., 2018, BIORXIV
[6]   Rapid geometric feature signaling in the simulated spiking activity of a complete population of tactile nerve fibers [J].
Delhaye, Benoit P. ;
Xia, Xinyue ;
Bensmaia, Sliman J. .
JOURNAL OF NEUROPHYSIOLOGY, 2019, 121 (06) :2071-2082
[7]  
Gerstner W., 2002, Spiking Neuron Models: Single Neurons, Populations
[8]   A fast learning algorithm for deep belief nets [J].
Hinton, Geoffrey E. ;
Osindero, Simon ;
Teh, Yee-Whye .
NEURAL COMPUTATION, 2006, 18 (07) :1527-1554
[9]  
Nguyen HG, 2018, IEEE INT C ENG COMP, P1, DOI [10.1109/ICECCS2018.2018.00009, 10.1109/ICOPS35962.2018.9575287]
[10]   A QUANTITATIVE DESCRIPTION OF MEMBRANE CURRENT AND ITS APPLICATION TO CONDUCTION AND EXCITATION IN NERVE [J].
HODGKIN, AL ;
HUXLEY, AF .
JOURNAL OF PHYSIOLOGY-LONDON, 1952, 117 (04) :500-544