Large-area Tomographic Tactile Skin with Air Pressure Sensing for Improved Force Estimation

Haofeng Chen, Bedrich Himmel, Jiri Kubik, Matej Hoffmann, Hyosang Lee · 2025

Read Large-area Tomographic Tactile Skin with Air Pressure Sensing for Improved Force Estimation by Haofeng Chen, Bedrich Himmel, Jiri Kubik, Matej Hoffmann, Hyosang Lee (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-13036.txt.

Introduction

Haofeng Chen and co-authors build a large touch skin that reads force more truly in a 2025 paper. Skins based on charge maps can find touch spots with few wires, but their feel fades by zone. That fade caps how well they can gauge force at each spot.

The new skin adds an air press layer that gives one steady force read for the whole patch. A one-sit tune learns smooth gain and shift maps by zone from both layers. With that tune, touch force reads stay even across the skin for many spots at once.

Catalog record

CREATORHaofeng Chen, Bedrich Himmel, Jiri Kubik, Matej Hoffmann, Hyosang Lee
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE26 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-13036

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Large-area Tomographic Tactile Skin with Air Pressure Sensing for Improved Force Estimation Haofeng Chen, Bedrich Himmel, Jiri Kubik, Matej Hoffmann, Hyosang Lee Abstract This paper presents a dual-channel tactile skin that integrates Electrical Impedance Tomography (EIT) with air pressure sensing to achieve accurate multi-contact force detection. The EIT layer provides spatial contact information, while the air pressure sensor delivers precise total force measurement. Our framework combines these complementary modalities through: deep learning-based EIT image reconstruction, contact area segmentation, and force allocation based on relative conductivity intensities from EIT. The experiments demonstrated 15.1% average force estimation error in single-contact scenarios and 20.1% in multi-contact scenarios without extensive calibration data requirements. This approach effectively addresses the challenge of simultaneously localizing and quantifying multiple contact forces without requiring complex external calibration setups, paving the way for practical and scalable soft robotic skin applications. I INTRODUCTION Large-area tactile sensors have emerged as a promising research topic due to their potential to enhance both robot’s and human user’s safety in dynamic environments [ 1 , 2 , 3 , 4 ] . Seamlessly covering a robot’s complex and three-dimensional body surfaces is required to perceive contact positions and forces over the complete body. To handle this nontrivial problem, multiple arrays of sensors have been commonly suggested [ 5 , 6 ] . However, these approaches deploy a large number of sensing modules, which can increase manufacturing costs and compromise the sensors’ flexibility and stretchability. To address these limitations, reconstruction-based methods such as electrical impedance tomography (EIT) have emerged as a promising alternative. EIT-based tactile sensors reconstruct pressure-induced conductivity distributions of a large sensing area using a few electrodes. This approach offers a stretchable, cost-effective, and feasible solution for large-area tactile sensing [ 7 , 8 , 9 ] . Despite these advantages, EIT-based tactile sensors have few challenges related to the non-uniform sensitivity distribution [ 10 ] , which is mainly caused by the electric field’s electrode placement dependency [ 11 ] and measurement noise and model inaccuracy susceptibility [ 12 ] . As the non-uniform sensitivity directly impacts the sensor’s contact force and location estimation, it has been studied by several researchers by using reliable ground-truth contact data for improving physics-based model calibration or as training data for machine learning. Lee et al. [ 13 ] introduced a single-point indentation testbed to obtain a coefficient field. This coefficient field was then used to calibrate the conductivity image into a pressure field, ensuring approximately constant sensitivity across all locations and thereby improving force detection accuracy. Later, they developed a sophisticated multi-point indenter and sim-to-real transfer learning pipeline for multi-contact prediction [ 12 ] . Chen and Liu leveraged the symmetry of the sensor structure [ 10 ] . They proposed the intensity scaling method to correct reconstructed conductivity magnitudes, achieving a more uniform sensitivity distribution. Similarly, Chen et al. [ 14 ] introduced a pseudo-array method based on single-contact experiment data. This approach equally divided the EIT-based sensor’s surface into a pseudo-array unit, obtained a force-to-conductivity mapping relationship for each unit, and developed the Jacobian vector correction method, which enabled near-uniform sensitivity across the sensing area without extensive calibration data [ 15 ] ; however, their work was limited to single-point contact force detection. Fig. 1: Overview of the proposed dual-channel tactile skin incorporating electrical impedance tomography (EIT)-based tactile sensing layer and air pressure sensing layer. All the aforementioned EIT-based tactile sensors require additional calibration apparatus to obtain ground truth contact locations and force magnitudes, such as an XYZ stage or a motion capture device with a loadcell, even multi-point indenter systems. The use of these apparatuses hinders the real-world application of the above methods because the contact experiment is cumbersome, time-consuming, and difficult to scale, especially when the robot skin consistently deforms or is installed on complex robot body surfaces where data collection system cannot reach. Integrating additional modalities or force measurement channels can be a good option to avoid this calibration challenge. Chen et al. [ 16 ] introduced a multimodal EIT sensor system with two stacked EIT layers: one layer with a spacer mesh to control pressure estimation sensitivity, and another layer for the structural modality sensing, which provides contact geometric information. In this way, one sensor could complement another. Although this approach is inspiring, it still requires two layers of EIT sensors that complicate sensor design. As a practical alternative, a soft pneumatic sensor introduced by Park et al. [ 17 ] can be used to provide nearly ground-truth contact force while preventing design complications. Reliable sensing of pressure or force magnitude is key in some application areas, such as safe human-robot collaboration. Air pressure sensors like AIRSKIN are being deployed on industrial robots, making them collaborative (see [ 18 ] for an empirical assessment of the effects), and it is possible to acquire the necessary safety ratings. In this work, we propose a multi-channel sensor system that integrates an air pressure sensing layer into an EIT sensor, as shown in Figure 1 and its reconstruction framework to enhance contact force estimation. This hybrid design leverages the complementary strengths of both sensing modalities: the EIT sensor offers the possibility of multiple contact localization, and the soft pneumatic sensor provides accurate total external contact force estimation, replacing impractical external calibration setups. II Sensor System Design The design of the EIT sensing layer and air pressure sensing layer are mostly adopted from Chen et al.’s work [ 19 ] and Park et al.’s work [ 17 ] , respectively. Since the details about each sensor’s fabrication, electronics, material characteristics, and signal processing can be found in the literature, this section focuses on the integration of the two sensors in hardware design. II-A Dual-channel Tactile Skin In order to make the dual-channel tactile skin, the air-pressure sensing layer and the EIT layer are fabricated separately. The EIT layer is then integrated into the air-pressure sensing layer that has a slot for encapsulation, as illustrated in the Figure 2 . Fig. 2: Fabrication process of the flexible EIT-based and air-pressure-based dual-modal tactile sensor. II-A 1 EIT layer The EIT layer, which estimates pressure-induced conductivity distribution, is made of a porous structure and ionic liquid [ 19 ] . The porous structure is polyurethane (PU) foam made by the foaming method. On the boundary of the PU foam, sixteen electrodes that are made of silver tape are attached. These electrodes are used to inject an alternating current of 40 kHz and measure voltages. The PU foam has a square shape with a length of 100 mm and a thickness of 5 mm, and the electrodes have a 5 mm length with 25 mm apart one another. After the PU foam is made, ionic liquid is poured into the foam. The ionic liquid is saline water with 125-300 \mu S/cm conductivity. To prevent evaporation of the liquid, an additional elastomer (Ecoflex 00-30, Smooth-On, USA) is then placed to seal the PU foam. The local impedance change is measured by the well-established EIT technique. A customized EIT acquisition board is utilized, featuring 16 electrodes, 4 analog multiplexers (ADG706), an AD5940 impedance measurement integrated circuit (IC), and an STM32 microcontroller unit (MCU). The AD5940 is responsible for current injection

Read the complete text