Data-efficient Tactile Sensing with Electrical Impedance Tomography

TL;DR

The paper tackles data scarcity in EIT-based tactile sensing, where the inverse problem $\Delta \mathbf{V} = \boldsymbol{J} \Delta \boldsymbol{\sigma}$ complicates high-resolution reconstruction. It proposes a data-augmentation pipeline that converts a single EIT frame into 32 distinct readouts by exploiting circular sensor symmetry through rotation and flipping, thereby enriching training data and covering unobserved contact positions. A DNN with an MLP encoder and a CNN decoder learns to map augmented measurements to tactile conductivity maps, trained on the EdEIT simulated dataset and validated on a real-flexible sensor. Results show CC improvements up to $17.35\%$ and RE reductions up to $23.74\%$, with a data-efficiency gain of over $31\times$ fewer raw measurements, enabling robust reconstruction under noise and broad applicability to EIT-based tactile sensing.

Abstract

Electrical Impedance Tomography (EIT)-inspired tactile sensors are gaining attention in robotic tactile sensing due to their cost-effectiveness, safety, and scalability with sparse electrode configurations. This paper presents a data augmentation strategy for learning-based tactile reconstruction that amplifies the original single-frame signal measurement into 32 distinct, effective signal data for training. This approach supplements uncollected conditions of position information, resulting in more accurate and high-resolution tactile reconstructions. Data augmentation for EIT significantly reduces the required EIT measurements and achieves promising performance with even limited samples. Simulation results show that the proposed method improves the correlation coefficient by over 12% and reduces the relative error by over 21% under various noise levels. Furthermore, we demonstrate that a standard deep neural network (DNN) utilizing the proposed data augmentation reduces the required data down to 1/31 while achieving a similar tactile reconstruction quality. Real-world tests further validate the approach's effectiveness on a flexible EIT-based tactile sensor. These results could help address the challenge of training tactile sensing networks with limited available measurements, improving the accuracy and applicability of EIT-based tactile sensing systems.

Figures (10)

• Figure 1: Electrical Impedance Maps (EIMs). (a) The EIM of 104 measurements. (b) The EIM of 208 measurements.
• Figure 2: Illustration of the proposed data augmentation strategy. (a) - (c) Different positions of touch represented through rotational and flip transformations, demonstrating the rotation and flip relationship between (a), (b), and (c). (d) - (f) The corresponding EIM of different positions touch. The blue part shows 104 original EIT measurements. (d) is derived by shifting the rows of (e) upwards, and (f) is derived by flipping (e) along the diagonal.
• Figure 3: Two examples using the proposed data augmentation approach. (a) Rotation transformations are applied to the original data. (b) Flip transformations are applied to (a).
• Figure 4: The architecture of the DNN model for tactile reconstruction.
• Figure 5: Training and validation loss.