Cloth Simulation

Computer simulation of how cloth stretches, bends, wrinkles, and collides — essential for research on manipulating clothing.

Cloth simulation is a category of soft-body simulation that computer graphics has studied since the 1980s. Early geometric methods (such as Weil 1986) approximated wrinkle shapes using catenary curves and ignored dynamics entirely; physics-based methods model cloth as a mesh of point masses connected by springs, accounting for stretch, shear, bending, and gravity; more refined approaches use energy models or the finite element method. The difficulty is that cloth is very thin and extremely prone to self-collision and interpenetration, while being stiff along its stretch direction, so even a moderately large timestep can trigger numerical instability. Implementations commonly used in robotics research include the particle-based NVIDIA FleX and MuJoCo's flex deformable bodies (which can represent rope, cloth, and volumetric soft bodies). Real cloth's physical parameters are hard to measure accurately, so the sim-to-real gap for cloth tasks is usually larger than for rigid-body tasks.

ExampleSoftGym (CoRL 2020), built on NVIDIA FleX, provides tasks such as flattening a cloth, folding it in half, and letting it settle flat on the ground, used to test reinforcement learning on deformable-object manipulation.

Also called

Fabric Simulation

Related

Deformable-Body Simulation · Garment Manipulation · Deformable Object Manipulation · Position-Based Dynamics · Finite Element Method · SoftGym: Benchmarking Deep Reinforcement Learning for Deformable Object Manipulation

Sources

Wikipedia: Cloth modeling
SoftGym: Benchmarking Deep Reinforcement Learning for Deformable Object Manipulation (arXiv 2011.07215)
MuJoCo Documentation: Modeling - Deformable objects