Explicit friction modeling improves cloth simulation with neural networks
Explicit Frictional Contact Modeling in Neural Garment Simulation
Computer Vision and Pattern RecognitionGraphicsMachine Learning
Summary
Simulating how clothes move and stick to bodies is tricky because friction, which makes clothes stick or slide, is hard to control in AI-based simulators. The authors found that current neural networks don't respond well when trying to change the friction settings during use. They created a new method that adjusts friction after the AI predicts the garment's motion, without needing to retrain the model. This approach mimics real friction behavior better and runs faster than traditional methods.
What this means in practice
- •For computer graphics developers: Enhance cloth simulation tools with adjustable friction effects for more realistic virtual garments without retraining AI models.
- •For animation studios: Produce more accurate and visually consistent animations of clothing that respond naturally to different friction settings.
Authors
Thanh-Nguyen Truong, Vu-Hoang Tran, Ching-Chun Huang
Abstract
Graph neural networks (GNNs) simulate garments efficiently through physics-based self-supervision, but leave body-garment friction implicit or inside training energies, with no friction coefficient $μ$ to set at inference; retraining the tested GNNs with a coefficient-specific friction term barely responds to $μ$. Optimization-based solvers decide each contact's stick-slip state from the reaction that governs it, but only by re-optimizing the garment state at every step; a friction term in a one-pass training objective acts as a fixed-weight penalty on sliding, and the trained weights apply no per-contact Coulomb condition at inference. We therefore resolve friction explicitly after prediction: a Modular Contact Operator, attached to a frozen predictor without retraining or an iterative solve, estimates the normal load, applies a closed-form Coulomb correction along a path across body triangles, and propagates the induced motion through the cloth as a linearized membrane wave. On HOOD and ContourCraft, it raises sticking and lowers slip monotonically in $μ$, tracking a reference solver's response while coefficient-specific retraining stays flat, with low penetration and reference-consistent stretching at an eighth of the solver's runtime.