Robot Skin with Touch and Bend Sensing using Electrical Impedance Tomography
Haofeng Chen, Bin Li, Bedrich Himmel, Xiaojie Wang, Matej Hoffmann · 2025
Read Robot Skin with Touch and Bend Sensing using Electrical Impedance Tomography by Haofeng Chen, Bin Li, Bedrich Himmel, Xiaojie Wang, Matej Hoffmann (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-2503-13048.txt.
Introduction
Haofeng Chen and co-authors build a robot skin that can feel both touch and bending at once. Their sensor uses electrical impedance tomography, a method that infers what is happening across a surface from electrical readings at its edges, needing little wiring. Bending is the hard part: it shifts the baseline readings and mixes with the signals from touch. The team pairs a neural network that classifies the surface's state with an adaptive reference scheme that separates the two effects, plus a regression model that turns readings into bending angles.
The skin itself is a magnetic hydrogel sheet that flexes with the surface beneath it. In tests, the system estimated bending angles well, told touch, bending, and rest apart reliably, and located touches far better under bending than fixed-reference methods. Live trials confirmed it can track several touches and changing angles at the same time.
Catalog record
| CREATOR | Haofeng Chen, Bin Li, Bedrich Himmel, Xiaojie Wang, Matej Hoffmann |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 26 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-13048
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Robot Skin with Touch and Bend Sensing using Electrical Impedance Tomography Haofeng Chen, Bin Li, Bedrich Himmel, Xiaojie Wang, Matej Hoffmann Abstract Flexible electronic skins that simultaneously sense touch and bend are desired in several application areas, such as to cover articulated robot structures. This paper introduces a flexible tactile sensor based on Electrical Impedance Tomography (EIT), capable of simultaneously detecting and measuring contact forces and flexion of the sensor. The sensor integrates a magnetic hydrogel composite and utilizes EIT to reconstruct internal conductivity distributions. Real-time estimation is achieved through the one-step Gauss-Newton method, which dynamically updates reference voltages to accommodate sensor deformation. A convolutional neural network is employed to classify interactions, distinguishing between touch, bending, and idle states using pre-reconstructed images. Experimental results demonstrate an average touch localization error of 5.4 mm (SD 2.2 mm) and average bending angle estimation errors of 1.9° (SD 1.6°). The proposed adaptive reference method effectively distinguishes between single- and multi-touch scenarios while compensating for deformation effects. This makes the sensor a promising solution for multimodal sensing in robotics and human-robot collaboration. I Introduction The importance of endowing robots with touch has been recognized for several decades and a large number of technologies have been developed (e.g., [ 1 , 2 ] for surveys). The focus has largely been on tactile sensing for manipulation, as equipping robot fingers or hands with tactile sensors requires only relatively small patches of electronic skin. Whole-body artificial skins for robots have been an exception, with only few successful solutions such as [ 3 , 4 ] . The potential of sensing over the whole body surface has also not been fully explored. Cheng et al. provide an overview of applications of their electronic skin in [ 5 ] . Even for the so-called whole-body skins [ 5 , 6 ] , it is not the complete surface of the robot that is covered. In particular the areas around the robot joints are typically without skin coverage. Complete coverage would be desired for both physical and social Human-Robot Interaction [ 7 , 8 ] . The main challenge for articulated robots lies in sustaining large deformation and stretch while still delivering touch information. While measuring the angle of deformation (bending) may not be key for classical robots, which have accurate joint encoders, bending information may be relevant in soft robots, providing valuable information about their body configuration (see e.g., [ 9 ] ). While soft and flexible tactile sensors have been explored, existing solutions often lack the ability to simultaneously detect and sense touch and bending. For example, Kramer et al. [ 10 ] developed a soft curvature sensor for joint angle tracking, but it cannot sense touch. Noda et al. [ 11 ] introduced a stretchable liquid-based tactile sensor with a dual-channel design, but its low spatial resolution limits multi-touch sensing. Totaro et al. [ 12 ] proposed a multimodal mechanosensing system using resistive and optical components, capable of distinguishing between pressure and bending. However, it only supports single-point touch sensing. Zhang et al. [ 13 ] designed a high-sensitivity microwave-based sensor capable of sensing both pressure and bending, but its reliance on electromagnetic coupling restricts it to small, localized areas. These limitations highlight the need for a solution that integrates touch and deformation measurements with high spatial resolution and scalability. Fig. 1: Overview of EIT-based flexible tactile sensor for simultaneous touch and bending sensing. Electrical Impedance Tomography (EIT) presents a promising alternative for distributed sensing without requiring multiple discrete sensor units, making it well suited for whole-body robotic applications. Prior research has explored EIT for touch sensing, such as Kato et al. [ 14 ] , who developed a soft, pressure-sensitive EIT-based tactile sensor for finger touch sensing. Similarly, and Park et al. [ 15 ] , who introduced a large-area, face-shaped EIT sensor for multi-touch recognition. EIT has also been applied to shape sensing in soft robots, with Avery et al. [ 16 ] and Xin et al. [ 17 ] using it to map deformations in real time. However, these approaches either lack touch sensing or require complex hardware, limiting their practicality for real-time applications. Dong et al. [ 18 ] demonstrated 3D-scanner-assisted EIT for tactile interaction mapping on deformable surfaces, but their approach required complex hardware and was not suitable for real-time applications. Despite these advancements, few studies have addressed simultaneous sensing of touch and bending in large-deformation robotic applications using EIT. The primary challenge in achieving simultaneous touch and bending sensing using EIT lies in signal coupling—both touch and bending alter the impedance distribution in a complex manner, making it difficult to separate their effects. Additionally, EIT is an ill-posed inverse problem, where small measurement variations can lead to significant reconstruction errors, especially under large deformations. In this paper, we present a dual-modal EIT-based flexible tactile sensor, capable of simultaneously sensing both external touch and bending of the sensor. As shown in Fig. 1, the sensor integrates EIT into the robotic arm, enabling it to track bending angles and localize touch positions through conductivity distribution analysis. EIT reconstructs internal conductivity patterns by applying electrical currents and measuring boundary voltages via surface electrodes. Touch interactions generate localized conductivity peaks, whereas bending induces distributed conductivity variations. To address signal coupling, our system employs a dynamic reference framework: (1) a pre-reconstruction step estimates initial conductivity distribution, (2) a classification step distinguishes between touch and bending, (3) bending-induced variations update the reference dynamically, and (4) touch sensing is refined using the updated reference, ensuring precise localization. II METHODS AND SYSTEM DESIGN II-A Working Principle of EIT for Multimodal Sensing The EIT reconstruction problem maps the internal conductivity distribution of a conductive domain using known excitation currents and measured boundary voltages. This study employs time-difference EIT, a technique focused on visualizing conductivity variations between two observed states [ 19 ] . In difference imaging, the connection between the change in conductivity, \Delta\bm{\sigma}\in{\mathbb{R}^{N}} , and the change in measured voltage, \Delta\mathbf{V}\in{\mathbb{R}^{M}} , can be approximated as a linear relationship: \Delta\mathbf{V}={\text{ }}\mathbf{J}\Delta\bm{\sigma}{\text{ + }}{\mathbf{n}_{noise}} (1) where the sensitivity matrix \mathbf{J}{\text{ }}\in{\mathbb{R}^{M\times N}}{\text{ }}(M\ll N) , also known as the Jacobian matrix, is computed using the finite element method (FEM). Each element of \mathbf{J} is given by {J_{mn}}={\text{ }}{\left.{\frac{{\partial\Delta{v_{m}}}}{{\partial\Delta{\sigma_{n}}}}}\right|_{{\sigma_{0}}}} , representing the sensitivity of voltage changes to conductivity variations. Additionally, {\mathbf{n}_{noise}}\in{\mathbb{R}^{M}} accounts for measurement noise. We define the conductivity change vector as \Delta\bm{\sigma}{\text{ = }}\left[{\Delta{\sigma_{1}},\Delta{\sigma_{2}},...,\Delta{\sigma_{N}}}\right] , where \Delta{\sigma_{i}}={\text{ }}{\sigma_{i}}-{\sigma_{0}} . Here, {\sigma_{0}} is the background conductivity, and {\sigma_{i}} is the conductivity of the i -th element. Similarly, the voltage change vector \Delta\mathbf{V}{\text{ = }}\left[{\Delta{v_{1}},\Delta{v_{2}},...,\Delta{v_{M}}}\right] is calculated as \Delta{v_{i}}{\text{ = }}{v_{i}}{\text{ }}-{\text{ }}v_{i}^{ref} , where {\mathbf{V}^{ref}}{\text{ = }}\left[{v_{1}^{ref},v_{{}_{2}}^{ref},...,v_{M}^{ref}}\right] is the reference voltage vector and \mathbf{V}{\text{ = }}\left[{{v_{1}},{v_{2}},...,{v_{M}}}\right] is the current voltage vector. In this work, the one-step Gauss-Newton method is employed to solve the EIT inverse problem due to its ability to deliver fast, real-time estimation. By leveraging time-difference imaging [ 20 ] , the relationship between the reconstructed conductivity changes \Delta\bm{\sigma} and the measured boundary voltage variations \Delta\mathbf{V} can be expressed as follows: \Delta\bm{\sigma}{\text{ }}={\text{ }}{({\mathbf{J}^{T}}\mathbf{J}+{\lambda^{2}}\mathbf{R})^{-1}}{\mathbf{J}^{T}}\Delta\mathbf{V} (2) where, \mathbf{J}^{T} represents the transpose of the Jacobian matrix \mathbf{J} , \lambda is a scalar hyperparameter, which is set to 0.2 to balance image