Smoothed Particle Hydrodynamics

A meshless simulation method that breaks a fluid into particles and computes motion by weighting over nearby particles.

Smoothed Particle Hydrodynamics was proposed by Gingold, Monaghan, and Lucy in 1977, originally to model galaxy and star formation in astrophysics. Rather than dividing space into a grid, it treats a fluid as a swarm of particles that move with the material (a Lagrangian method, meaning it follows material points rather than fixed locations): each particle's density, pressure, and other physical quantities are computed as a weighted sum over neighboring particles within a “smoothing length,” using a kernel function, and the resulting forces then drive the particle forward. Because it needs no mesh, it is naturally suited to free surfaces, splashing, pouring, and other flows with drastic shape changes, and it is also widely used for fluid effects in computer graphics and games. In embodied AI it is used when a robot needs to learn to pour water or scoop liquid — the Genesis simulator, for example, has a built-in SPH solver. The cost is that computation grows heavy as particle counts rise, and precisely maintaining a liquid's incompressibility is not easy.

ExampleIn Genesis, represent the water in a cup as SPH particles and train a robot arm to pour it into another cup, penalizing the policy for how many particles spill out.

Also called

SPH

Related

Fluid Simulation · Position-Based Dynamics · Material Point Method · Finite Element Method · Genesis · Deformable-Body Simulation

Sources

Wikipedia - Smoothed-particle hydrodynamics
GitHub - Genesis-Embodied-AI/Genesis