[Paper Review] Elastic Context: Encoding Elasticity for Data-driven Models of Textiles

Alberta Longhini, Marco Moletta+7 more|arXiv (Cornell University)|Sep 12, 2022

Textile materials and evaluations4 citations

TL;DR

This paper introduces Elastic Context (EC), a data-driven method to encode textile elasticity using stress-strain curve-inspired embeddings for robotic manipulation. By integrating EC into Graph Neural Networks (GNNs), the model achieves accurate force prediction across diverse elastic behaviors in simulation and real-world experiments, with EC dimensionality significantly improving non-linear force modeling in real scenarios.

ABSTRACT

Physical interaction with textiles, such as assistive dressing, relies on advanced dextreous capabilities. The underlying complexity in textile behavior when being pulled and stretched, is due to both the yarn material properties and the textile construction technique. Today, there are no commonly adopted and annotated datasets on which the various interaction or property identification methods are assessed. One important property that affects the interaction is material elasticity that results from both the yarn material and construction technique: these two are intertwined and, if not known a-priori, almost impossible to identify through sensing commonly available on robotic platforms. We introduce Elastic Context (EC), a concept that integrates various properties that affect elastic behavior, to enable a more effective physical interaction with textiles. The definition of EC relies on stress/strain curves commonly used in textile engineering, which we reformulated for robotic applications. We employ EC using Graph Neural Network (GNN) to learn generalized elastic behaviors of textiles. Furthermore, we explore the effect the dimension of the EC has on accurate force modeling of non-linear real-world elastic behaviors, highlighting the challenges of current robotic setups to sense textile properties.

Motivation & Objective

• Address the challenge of modeling complex, non-linear elastic behaviors in textiles for robotic manipulation, where material and construction properties are often unknown a-priori.
• Overcome limitations of traditional analytical models that are computationally expensive and not real-time applicable.
• Develop a generalizable, data-driven encoding of elasticity that can be used with common robotic sensing hardware.
• Investigate how EC dimensionality affects modeling accuracy for both linear and non-linear elastic behaviors in simulation and real-world settings.
• Enable robust force forecasting in assistive robotics and human-robot interaction tasks requiring precise force control on textiles.

Proposed method

• Define Elastic Context (EC) as a stress-strain curve-inspired embedding derived from force-torque measurements during robotic pulling and stretching of textiles.
• Reformulate traditional textile engineering stress-strain curves for robotic applications, enabling real-time encoding of elastic behavior.
• Integrate EC as a conditional input to Graph Neural Networks (GNNs) to model force dynamics and deformation of textile nodes under robot actions.
• Use PyBullet simulations to generate synthetic training data with diverse elastic properties for initial evaluation.
• Collect real-world interaction data using two Franka-Emika Panda robots to evaluate EC performance on non-linear elastic behaviors.
• Train multi-layer perceptrons (MLPs) with varying EC dimensions (n_EC ∈ [0,5]) on real-world data to predict next-step forces based on actions and EC.

Experimental results

Research questions

• RQ1 Can Elastic Context (EC) effectively encode the elastic behavior of textiles in a way that improves data-driven force modeling for robotic manipulation?
• RQ2 How does the dimensionality of EC affect the accuracy of force prediction in both simulated and real-world scenarios?
• RQ3 Does EC outperform baseline models that do not incorporate elastic information when modeling linear and non-linear elastic behaviors?
• RQ4 Can EC generalize to unseen textile samples with diverse elastic properties in real-world robotic interactions?
• RQ5 Is EC effective in capturing non-linear force dynamics in real-world textile manipulation, especially when compared to low-dimensional EC approximations?

Key findings

• The GNN with EC (n_EC = 1) achieves force prediction error (0.269 ± 0.301 N) comparable to the oracle model (0.207 ± 0.239 N), demonstrating EC's effectiveness in encoding elasticity.
• The baseline model without EC shows significantly higher force prediction error (3.169 ± 2.434 N), confirming that elastic information is crucial for accurate modeling.
• In real-world experiments, increasing EC dimensionality from 1 to 5 reduces force prediction MSE, with the lowest error achieved at higher dimensions, indicating improved modeling of non-linear behaviors.
• The model with n_EC = 1 performs poorly on non-linear real-world force profiles, while higher-dimensional EC variants show substantial improvement, highlighting the importance of EC dimensionality for complex dynamics.
• Graph rollout predictions remain comparable across models due to isotropic elasticity and displacement-based actions, but EC's impact is most evident in force forecasting.
• EC enables accurate force modeling without requiring prior knowledge of yarn material or construction technique, making it suitable for real-world robotic deployment.