Roughness Encoding for Discrimination of Surfaces in Artificial Active-Touch

Abstract

A 2 × 2 array of four microelectromechanical system (MEMS) tactile microsensors is integrated with readout electronics in the distal phalanx of an anthropomorphic robotic finger. A total of 16 sensing elements are available in a 22.3-mm area (i.e., 72 units/cm ) of the artificial finger, thus achieving a density comparable with human Merkel mechanoreceptors. The MEMS array is covered by a polymeric packaging with biomimetic fingerprints enhancing the sensitivity in roughness encoding. This paper shows the ability of the sensor array to encode roughness for discrimination of surfaces, without requiring dedicated proprioceptive sensors for end-effector velocity. Three fine surfaces with 400-, 440-, and 480- μm spatial periods are quantitatively evaluated. Core experiments consisted in active-touch exploration of surfaces by the finger executing a stereotyped human-like movement. A time-frequency analysis on pairs of tactile array outputs shows a clustering of the fundamental frequency, thus yielding 97.6% worst-case discrimination accuracy with a k -nearest-neighbor (k-NN) classifier. Hence, surfaces differing down to 40 μm are identified in active-touch by both hardware and processing methods based on exteroceptive tactile information. Finally, active-touch results with five textiles (which differ in texture or orientation) are shown as a preliminary qualitative assessment of discrimination in a more realistic tactile-stimulation scenario.