On-Body Force Sensing Using a Soft E-Textile Matrix: Development, Modeling, and Evaluation

Abstract

Accurate force estimation on the human body is inherently challenging due to variations in body geometry, surface curvature, and soft tissue deformation, which introduce complexities in sensor response. In this article, we present the development and evaluation of a low-profile, textile-based, and biocompatible flexible sensor matrix composed of 72 sensing cells ( $6\times 12$ configurations) with a spatial fine resolution of 1 cm², designed for wearable applications requiring precise on-body force measurement. The goal of this article is to improve force-estimation accuracy for applications such as medical monitoring, rehabilitation, and human-computer interaction. This study contributes a fully soft e-textile matrix together with a matching data-driven modeling framework. We investigated multiple modeling techniques, including exponential and polynomial regression, as well as neural-network architectures such as sequential multilayer perceptron (MLP) and long short-term memory (LSTM) networks. The LSTM model outperformed others by capturing hysteresis effects and improving force-estimation accuracy. The sensor matrix was first validated using balance weights on flat surfaces, achieving a root mean square error (RMSE) of 0.02 N. It was, then, evaluated on soft, curved mannequin surfaces representing varying geometries—low (torso), medium (thigh), and high (forearm) curvatures—resulting in a minimum RMSE of 0.056 N, a maximum error of 0.14 N, and an average RMSE of 0.09 N. These results demonstrate the matrix’s accuracy and robustness for on-body force sensing on deformable, nonuniform surfaces.