Cloth Simulation is the computational modelling of textile deformation and dynamics, representing fabric as a mesh of particles connected by spring constraints (stretch, shear, and bend) or by a continuum-mechanics model, and integrating the equations of motion to produce plausible cloth behaviour under gravity, wind, collision, and user interaction. The particle-spring model, popularised by Provot (1995), remains prevalent in real-time applications; more accurate results for offline rendering use finite-element or position-based dynamics (PBD) methods. Collision detection and response against rigid bodies and self-collision are the principal computational bottlenecks, requiring spatial acceleration structures such as bounding volume hierarchies. Cloth simulation is a sub-discipline of physically based animation used in character clothing, flag animation, curtains, and virtual fashion design within game engines, VFX pipelines, and metaverse avatar systems.
Content
- The canonical cloth model represents a rectangular fabric swatch as a grid of point masses joined by three types of spring: structural springs along the grid axes that resist stretch, shear springs along diagonals that resist skewing, and flexion (bend) springs connecting second-nearest neighbours that provide stiffness against folding. At each time step forces are summed at each particle and positions are integrated using explicit Euler, Verlet, or implicit integration schemes. Implicit integration (Baraff and Witkin 1998) allows much larger time steps without instability at the cost of solving a sparse linear system each frame.
- Position-Based Dynamics (PBD), introduced by Müller et al. (2007), replaced force-based integration in many real-time contexts. Rather than computing forces, PBD directly adjusts positions to satisfy distance, bending, and collision constraints in an iterative projection loop. PBD is unconditionally stable for common constraint types and runs efficiently on GPUs, making it the preferred approach in game engines such as Unreal Engine (Chaos Cloth) and Unity (DOTS Physics). Extended Position-Based Dynamics (XPBD) adds compliance parameters that recover physically meaningful stiffness values.
- Collision handling remains the dominant performance challenge. Self-collision — cloth intersecting itself during billowing or wrinkling — requires all-pairs proximity queries mitigated by spatial hashing or continuous collision detection. Character cloth must also respond to skinned mesh deformation, handled via proxy collision shapes (capsules, spheres) around limbs or via direct GPU skinning feedback. Offline VFX pipelines (Houdini, Marvelous Designer) prioritise accuracy over frame rate and use finer meshes, iterative solvers, and full self-collision for film-quality results.
- In metaverse and virtual-fashion contexts, cloth simulation enables digital garment try-on, avatar personalisation, and procedural textile variation. Real-time cloth quality is increasingly achieved through machine-learning surrogate models that approximate offline simulation results at interactive rates, enabling high-fidelity avatars with dynamic clothing at scale.