Learning Simulatable Models of Cloth with Spatially-varying Constitutive Properties

2025-07-28

AI Summary

This work addresses two key challenges in cloth simulation: modeling spatially varying mechanical properties and the computational expense and membrane locking artifacts inherent in finite element methods. To this end, we propose Mass-Spring Net—a differentiable mass-spring framework that directly infers spatially varying material parameters from motion observation data via a dual-objective loss function combining force and impulse residuals. Our approach circumvents numerical artifacts of traditional solvers, requires no explicit physical modeling or PDE solving, and achieves significantly higher training efficiency than graph neural networks and neural ODEs. It delivers high-fidelity dynamic reconstruction across multi-source datasets and exhibits strong generalization to unseen scenarios. The core contribution is the first integration of a differentiable mass-spring model with kinematically consistent, physics-informed losses—enabling efficient, robust, and interpretable surrogate modeling of spatially heterogeneous cloth.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Learning with Manifolds
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Web Mining and Content Analysis: Web data generation and simulation
• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems

Abstract

Materials used in real clothing exhibit remarkable complexity and spatial variation due to common processes such as stitching, hemming, dyeing, printing, padding, and bonding. Simulating these materials, for instance using finite element methods, is often computationally demanding and slow. Worse, such methods can suffer from numerical artifacts called ``membrane locking'' that makes cloth appear artificially stiff. Here we propose a general framework, called Mass-Spring Net, for learning a simple yet efficient surrogate model that captures the effects of these complex materials using only motion observations. The cloth is discretized into a mass-spring network with unknown material parameters that are learned directly from the motion data, using a novel force-and-impulse loss function. Our approach demonstrates the ability to accurately model spatially varying material properties from a variety of data sources, and immunity to membrane locking which plagues FEM-based simulations. Compared to graph-based networks and neural ODE-based architectures, our method achieves significantly faster training times, higher reconstruction accuracy, and improved generalization to novel dynamic scenarios.

Problem

• Modeling spatially-varying cloth properties efficiently
• Overcoming membrane locking in cloth simulation
• Learning material parameters from motion data

Innovation

• Learns surrogate model from motion observations
• Uses force-and-impulse loss function
• Discretizes cloth into mass-spring network