Learning physics property parameters of fabrics and garments with a physics similarity neural network

Tools

Duan, Li and Boyd, Lewis and Aragon-Camarasa, Gerardo (2022) Learning physics property parameters of fabrics and garments with a physics similarity neural network. IEEE Access, 10. pp. 114725-114734. ISSN 2169-3536 (https://doi.org/10.1109/access.2022.3217458)

[thumbnail of Duan-etal-IEEE-Access-2022-Learning-physics-property-parameters-of-fabrics-and-garments]
Preview
 Text. Filename: Duan_etal_IEEE_Access_2022_Learning_physics_property_parameters_of_fabrics_and_garments.pdf
Final Published Version
License: Creative Commons Attribution 4.0 logo
Download (1MB)| Preview

Abstract

Predicting the physics properties of deformable objects such as garments and fabrics is a challenge in robotic research. Directly measuring their physics properties in a real environment is difficult Bouman et al. (2010). Therefore, learning and predicting the physics property parameters of garments and fabrics can be conducted in simulated environments. However, garments have collars, sleeves, pockets and buttons that change how garments deform and simulating these is time-consuming. Therefore, in this paper, we propose to predict the physics parameters of real fabrics and garments by learning the physics similarities between simulated fabrics via a Physics Similarity Network (PhySNet). For this, we estimate wind speeds generated by an electric fan and area weights to predict the bending stiffness parameters of real fabrics and garments. We found that PhySNet coupled with a Bayesian optimiser can predict physics property parameters and improve state-of-art by 34.0% for fabrics and 68.1% for garments.

ORCID iDs

Duan, Li, Boyd, Lewis ORCID logoORCID: https://orcid.org/0000-0003-4475-2223 and Aragon-Camarasa, Gerardo;
CORE (COnnecting REpositories)