Object Recognition Based on Hardness and Texture via Modified Force-Sensitive Fingertips of a Humanoid Hand

Abstract

Multimodal tactile perception offers a new opportunity for object recognition based on surface properties. Herein, we present a straightforward and low-cost approach to measuring hardness and texture via modified force-sensitive fingertips of a five-fingered robotic hand. Specifically, a rigid indenter and a glass bead are attached on the thumb and the index finger to enable hardness and texture perception, respectively. After data being processed with fast Fourier transform and principal component analysis, machine learning algorithms, including multilayer perceptron and support vector machines, are used to identify objects based on hardness and texture. Online object recognition demonstrates an accuracy of 86.0% in a seven-toy study. This work could provide a quick solution for object recognition using force sensors for humanoids.