A Sensor Position Localization Method for Flexible, Non-Uniform Capacitive Tactile Sensor Arrays

TL;DR

This work tackles sensor localization within flexible, non-uniform tactile skin arrays. It introduces VARSkin, a variable-density mutual-capacitance artificial skin, and a three-step localization pipeline (Point Log Map construction, interpolation, and threshold-based filtering) to predict every embedded sensor position without additional hardware. On two concealed-layout patches, the method achieves localization accuracy within ±2 mm and is supported by an error analysis detailing probe, data processing, and sampling factors. The approach enables scalable, adaptive tactile skins suitable for both coarse, large-area sensing and fine manipulation, with implications for robotic perception and manipulation tasks.

Abstract

Tactile sensing is used in robotics to obtain real-time feedback during physical interactions. Fine object manipulation is a robotic application that benefits from a high density of sensors to accurately estimate object pose, whereas a low sensing resolution is sufficient for collision detection. Introducing variable sensing resolution into a single tactile sensing array can increase the range of tactile use cases, but also invokes challenges in localizing internal sensor positions. In this work, we present a mutual capacitance sensor array with variable sensor density, VARSkin, along with a localization method that determines the position of each sensor in the non-uniform array. When tested on two distinct artificial skin patches with concealed sensor layouts, our method achieves a localization accuracy within $\pm 2mm$. We also provide a comprehensive error analysis, offering strategies for further precision improvement.

Figures (12)

• Figure 1: In this work, we propose a localization method to predict the location of each sensor in a non-uniform sensor density artificial skin. The method starts by constructing a point log map that shows a single sensor's recorded intensities at different probing locations. The point log map is then interpolated, filtered, and averaged, resulting in a single sensor location prediction. This method is then repeated for each sensor to compile the final sensor location predictions.
• Figure 2: The prevalent large-scale artificial skin implementation in the iCub robot has equal average sensor density across the robot arm. In contrast, the average biological sensor densities of human skin gradually increase from the base of the arm to the tip of the fingers. In our work, we aim to mimic the biological approach of variable sensor density which we see in humans. corniani2020tactilemaiolino2013flexible.
• Figure 3: We fabricated two $2.54$x$15.24cm$ skin sensor patches capable of real-time tactile perception. Top blue sensor - Patch A - has an even distribution of internal electrodes which is a predominant feature in state-of-the-art approaches. The bottom white and blue sensor - Patch B - has a variation of electrode locations along its length and demonstrates our novel approach to enable variable-density skin. Our contributed localization method generalizes to both configurations of electrodes.
• Figure 4: Our tactile skin patch detects touch by measuring changes in mutual capacitance, which is an electromagnetic field created by two overlapping electrodes. a) The electromagnetic field when a human finger is in close proximity to an electrode crossing point. b) A circuit diagram of the transmitter and receiver electrodes forming a mutual capacitor $C_{RT}$ in parallel with a human finger in close proximity.
• Figure 5: A probe was lightly pressed against a $5$x$20$ equally spaced grid of locations along the surface of the artificial skin. This figure shows the distances between the probe location and sensor locations plotted against the capacitance measurements of each sensor. The resulting trend is used in our novel localization method to predict the sensor locations in Patch B from a light touch.