Machine Learning-Enhanced Fingertip Tactile Sensing: From Contact Estimation to Reconstruction

Abstract

Fingertip tactile sensing is a crucial perceptual mechanism for both humans and robots, providing valuable data on contact detection, force feedback, and object properties such as surface texture, shape, and temperature. This paper presents a novel tactile sensing system that incorporates a six-axis force/torque sensor, combined with innovative methods to harness force and torque signals for enhanced tactile perception. The main contributions of this work are as follows: (1) we propose a new approach for determining contact positions and normals, identifying unintended movements, and analyzing surface texture features, (2) we introduce a technique for estimating an object's center of gravity (CoG), which is integrated with tactile sensing for more accurate object interaction, and (3) we integrate a machine learning framework using an artificial neural network to classify objects based on CoG estimations and tactile data. Finally, we demonstrate the effectiveness of the proposed system through object reconstruction examples, confirming its performance in real-world applications.