AirTac: A Contactless Digital Tactile Receptor for Detecting Material and Roughness via Terahertz Sensing

Abstract

Tactile sensing is an indispensable capability for humans or intelligent devices to engage in complex physical interactions. Mainstream tactile sensors are contact-based, which show limitation in measuring deformable objects and demanding high maintenance effort. As a complementary solution, a new paradigm of contactless tactile sensing is attracting much interest. While promising, they can only identify coarse-grained single tactile perception properties, either material type or surface roughness. In this paper, we propose airTac, which includes a novel contactless digital tactile receptor model, capable of simultaneously extracting these two basic tactile properties in a fine-grained resolution, by harnessing the massive bandwidth of terahertz(THz) frequency band. airTac is designed based on our key finding: while the impact of material type and surface roughness on THz signal intertwine with each other, they are actually separable, i.e., roughness manifests a certain pattern in distorting relative high-frequency components of the whole THz spectrum. Therefore, we custom-design a bio-inspired deep neural network model to decouple the intertwined THz signal, and distill the tactile perception properties embedded underlying the signal. We prototype airTac using a THz time domain spectroscopy, and perform extensive evaluation over 9 different daily material types with 39 different surface roughness. airTac achieves a material identification accuracy of 97.43% and a roughness classification accuracy of 91.46%, which demonstrates that airTac can identify common materials in daily life and exceed a fingertip spatial resolution of 1mm.