Neuromorphic approach to tactile edge orientation estimation using spatiotemporal similarity

Deepesh Kumar, Rohan Ghosh, Andrei Nakagawa-Silva, Alcimar B. Soares, Nitish V. Thakor

Research output: Contribution to journalArticle

Abstract

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 mechanoreceptors in the human hand offers a high spatiotemporal resolution to perceive the edge orientation quickly (<50 ms) and accurately (acuity of <3°). Inspired by the computational efficiency of biological tactile system, we present a neuromorphic approach to artificial tactile sensing that mimics the spike-based spatiotemporal 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 highlight the ability of the proposed method to efficiently leverage spatial and temporal information, by obtaining very precise orientation estimates (±1.67° error for edges oriented from 10° to 90°, with a step of 5°) in spite of a low-resolution sensor (169 mm2, 4×4 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°. This work highlights that even with a modest sensor size and resolution, spatiotemporal similarity metrics can be used to obtain very precise estimates of orientation. Such an approach has potential applications 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.

Original languageEnglish (US)
Pages (from-to)246-258
Number of pages13
JournalNeurocomputing
Volume407
DOIs
StatePublished - Sep 24 2020

Keywords

  • Biomimetic
  • Edge orientation estimation
  • Mechanoreceptors
  • Neuromorphic
  • Spatiotemporal
  • Tactile sensing

ASJC Scopus subject areas

  • Computer Science Applications
  • Cognitive Neuroscience
  • Artificial Intelligence

Fingerprint Dive into the research topics of 'Neuromorphic approach to tactile edge orientation estimation using spatiotemporal similarity'. Together they form a unique fingerprint.

  • Cite this