11th World Congress on Industrial Process Tomography
Enhanced Tactile Sensing with Electrical Impedance Tomography and Shallow Image Prior
H. Dong*, Z. Liu, D. Hu, Y. Yang
SMART Group, Institute for Digital Communications, School of Engineering, The University of Edinburgh, Edinburgh, UK
*Email: Huazhi.Dong@ed.ac.uk
ABSTRACT
The research on robot skin aims to augment robots' tactile perception by emulating human skin characteristics. Tactile sensing plays a pivotal role in the field of robotics, and hydrogels have displayed potential as tactile sensors owing to their supple and resilient properties. In this paper, we fabricated a haptic sensor utilizing hydrogels and proposed the Shallow Image Prior-based L1 regularization (SIP-L1) algorithm for enhancing tactile reconstruction. The proposed algorithm uses a three-layer Multi-Layer Perceptron (MLP) as a prior for the L1 regularized Electrical Impedance Tomography (EIT) inverse problem. The algorithm's performance was evaluated through simulations and experiments. Compared to the provided regularization algorithms, SIP-L1 exhibits evident advantages in terms of reconstruction quality enhancement, resulting in a decrease of 3.1% - 57.0% in the image error during numerical simulations, and an increase of 6.0%-789.4% in the MSSIM. The results obtained from both the simulations and practical experiments validate the superiority of this method, indicating promise for enhancing tactile perception in robots.
Keywords: electrical impedance tomography; hydrogel sensor; multi-layer perceptron; tactile sensing
Industrial Application: soft robotics, surgical robot
Sign-in to access the full text
Copyright © International Society for Industrial Process Tomography, 2023. All rights reserved.