Fluid Simulation

Numerically simulating how fluids like water or smoke flow and experience force inside a computer.

Fluid simulation means numerically approximating the equations that describe fluid motion (usually the Navier-Stokes equations) to compute how a liquid's or gas's velocity and pressure change over time. The branch chasing scientific precision is called computational fluid dynamics (CFD), used in engineering design for aircraft and piping; graphics and robot simulation instead prioritize speed and visual plausibility. Common approaches fall into three categories: grid-based methods (Eulerian, recording flow velocity on a fixed grid), particle-based methods (Lagrangian, such as smoothed-particle hydrodynamics, or SPH, which treats a fluid as a large swarm of particles), and hybrids of the two such as FLIP and the material point method. In embodied AI, tasks like pouring water, carrying a bowl of soup, or cleaning up a spill all need it; because of the large number of particles involved and the need to couple with the robot arm and container, the computational cost is far higher than rigid-body simulation, so a trade-off between accuracy and speed is usually necessary.

ExampleThe SoftGym benchmark's PourWater task (pouring all the water into a target cup) and TransportWater task (carrying a cup of water to a target location without spilling); Genesis simulates liquids with an SPH solver.

Also called

Liquid Simulation

Related

Smoothed Particle Hydrodynamics · Material Point Method · Position-Based Dynamics · Deformable-Body Simulation · SoftGym: Benchmarking Deep Reinforcement Learning for Deformable Object Manipulation · Genesis

Sources

Fluid animation - Wikipedia
SoftGym project page
Genesis Documentation: What is Genesis