A Machine Learning Approach to Contact Localization in Variable Density Three-Dimensional Tactile Artificial Skin

TL;DR

This work addresses contact localization on a curved, semi-conical artificial skin with non-uniform mutual-capacitance sensors by training a fully connected neural network to map raw sensor readings to 3D touch coordinates, bypassing the need to know the exact sensor placements. The method achieves $5.7 \pm 3.0$ mm localization accuracy and demonstrates robustness to variable sensor density, highlighting the practical value for robotics and prosthetics where geometry and sensor layouts are irregular. Key contributions include a complete fabrication/calibration/sensing pipeline, a compact neural localization model, and evidence that localization can succeed without explicit sensor geometry, enabling flexible tactile skins. The results suggest practical impact in reliable tactile sensing on complex surfaces, with future work extending to grid-based data collection and multi-touch scenarios.

Abstract

Estimating the location of contact is a primary function of artificial tactile sensing apparatuses that perceive the environment through touch. Existing contact localization methods use flat geometry and uniform sensor distributions as a simplifying assumption, limiting their ability to be used on 3D surfaces with variable density sensing arrays. This paper studies contact localization on an artificial skin embedded with mutual capacitance tactile sensors, arranged non-uniformly in an unknown distribution along a semi-conical 3D geometry. A fully connected neural network is trained to localize the touching points on the embedded tactile sensors. The studied online model achieves a localization error of $5.7 \pm 3.0$ mm. This research contributes a versatile tool and robust solution for contact localization that is ambiguous in shape and internal sensor distribution.

Figures (5)

• Figure 1: Contact localization model takes in a sensor image from any configuration of artificial tactile skin and determines the location of touch through a feedforward neural network.
• Figure 2: (a) CAD model for the curved geometry. (b) General wiring scheme. Sensors are located at each intersection of transmitter (TX) and receiver (RX) wires. (c) Fabricated skin sensor array where the sensing circuits are embedded within a layer of silicone.
• Figure 3: (a) Fully-connected neural network takes in sensor input of size $64$ and outputs a 3D coordinate. (b) Artificial tactile skin sends mutual capacitance measurements to an Arduino microcontroller that formats the readings and passes into the neural network.
• Figure 4: (a) The linear relationship between SNR and point logs dataset size suggests correlation. (b) Prediction error for contact localization models trained with varying sets of point logs.
• Figure 5: Our mutual capacitance sensor achieves spatial acuity consistent with sensing arrays of known distributions and human skin.