Deformable-Body Simulation

Physics simulation of objects that change shape under force, like cloth, rope, dough, or liquid.

Rigid-body simulation assumes an object's shape never changes, so a handful of numbers is enough to describe its pose; deformable-body simulation has to handle objects that change shape under force, with state made up of the positions of hundreds or thousands of nodes or particles, which is far more computationally and numerically demanding. Common methods include the finite element method (FEM, which cuts an object into a tetrahedral mesh and computes deformation from material parameters like Young's modulus and Poisson's ratio, well suited to elastic bodies), position-based dynamics (PBD, fast and stable, commonly used for cloth and rope), and the material point method (MPM, suited to large-deformation materials such as dough and sand). PhysX runs soft bodies with FEM on the GPU; Genesis integrates FEM, MPM, and PBD/SPH solvers together. This underlies research on manipulating flexible objects, such as folding laundry or kneading dough, though the gap to real materials is usually larger than for rigid-body simulation.

ExampleSoftGym (CoRL 2020), built on the particle simulator NVIDIA FleX, provides tasks such as flattening cloth, folding cloth, straightening rope, and pouring water, used to test how well reinforcement learning algorithms handle deformable objects.

Also called

soft-body simulation, flexible-body simulation

Related

Finite Element Method · Material Point Method · Position-Based Dynamics · Cloth Simulation · Deformable Object Manipulation · SoftGym: Benchmarking Deep Reinforcement Learning for Deformable Object Manipulation

Sources

NVIDIA PhysX 5 Documentation: Soft Bodies
SoftGym: Benchmarking Deep Reinforcement Learning for Deformable Object Manipulation (arXiv 2011.07215)
Genesis GitHub 仓库 (Chinese)