Learning to simulate complex physics
Learning to simulate complex physics
with graph networks
with graph networks
Embedded Files
Alvaro Sanchez-Gonzalez*, Jonathan Godwin*, Tobias Pfaff*, Rex Ying*,
Alvaro Sanchez-Gonzalez*, Jonathan Godwin*, Tobias Pfaff*, Rex Ying*,
Jure Leskovec, Peter Battaglia
Jure Leskovec, Peter Battaglia
Paper pre-print: arxiv.org/abs/2002.09405
Paper pre-print: arxiv.org/abs/2002.09405
teaser_3.mp4All experiments
All experiments
water_3d.mp4Water-3D
Water-3D
14k particles
800 steps
ground truth simulator: SPH
sand_3d.mp4Sand-3D
Sand-3D
19k particles
400 steps
ground truth simulator: MPM
goop_3d.mp4Goop-3D
Goop-3D
15k particles
300 steps
ground truth simulator: MPM
water_3d_s.mp4Water-3D-S
Water-3D-S
6k particles
800 steps
ground truth simulator: SPH
water.mp4Water
Water
2k particles
1000 steps
ground truth simulator: MPM
sand.mp4Sand
Sand
2k particles
320 steps
ground truth simulator: MPM
goop.mp4Goop
Goop
2k particles
400 steps
ground truth simulator: MPM
water_drop.mp4WaterDrop
WaterDrop
2k particles
1000 steps
ground truth simulator: MPM
water_drop_xl.mp4WaterDrop-XL
WaterDrop-XL
8k particles
1000 steps
ground truth simulator: MPM
randomfloor.mp4RandomFloor
RandomFloor
3.5k particles
600 steps
ground truth simulator: MPM
sandramps.mp4SandRamps
SandRamps
3.5k particles
400 steps
ground truth simulator: MPM
waterramps.mp4WaterRamps
WaterRamps
2.5k particles
600 steps
ground truth simulator: MPM
fluidshake.mp4FluidShake
FluidShake
1.4k particles
2000 steps
ground truth simulator: MPM
boxbath.mp4BoxBath
BoxBath
1k particles
150 steps
ground truth simulator: PBD
multi.mp4MultiMaterial
MultiMaterial
2k particles
1000 steps
ground truth simulator: MPM
continuous.mp4Continuous
Continuous
5k particles
400 steps
ground truth simulator: MPM
trained on: friction angle range [0, 30], [55-80]
inference: friction angle range [0, 90]
Generalization Experiments
Generalization Experiments
watervortex.mp4WaterVortex
WaterVortex
Generalization Experiment
Generalization Experiment
trained on WaterRamps (2.5k particles, 600 steps)
inference: 2x2 domain, 28k particles, 2500 steps
gen_ramps_small.mp4Ramps-Small
Ramps-Small
Generalization Experiment
Generalization Experiment
trained on WaterRamps (2.5k particles, 600 steps)
inference: 2x2 domain, 5k particles, 2000 steps
gen_ramps_large.mp4Ramps-Large
Ramps-Large
Generalization Experiment
Generalization Experiment
trained on WaterRamps (2.5k particles, 600 steps)
inference: 8x4 domain, 85k particles, 5000 steps
gen_ramps_sand.mp4Hourglass
Hourglass
Generalization Experiment
Generalization Experiment
trained on SandRamps (3.5k particles, 400 steps)
inference: 1x2 domain, 3.5k particles, 2000 steps
gen_multi.mp4MultiHippo
MultiHippo
Generalization Experiment
Generalization Experiment
trained on MultiMaterial (2k particles, 1000 steps)
inference: 1x1 domain, 4.5k particles, 2000 steps
Comparisons & Analysis
Comparisons & Analysis
boxbath_dpi.mp4Baseline Comparison: DPI
Baseline Comparison: DPI
BoxBath domain
While DPI uses hard-coded constraints to keep the box shape consistent, our model achieves this without any special treatment of the solid particles.
cconv_all.mp4Baseline Comparison: CConv
Baseline Comparison: CConv
Comparison in the following domains:
Water-3D-S (SPH)
BoxBath (PBD)
Sand, Water, Goop, MultiMaterial (MPM)
fail.mp4Examples of failure cases
Examples of failure cases
Above videos are indicative for our model's average performance. However, in our comprehensive experiments we have also found some interesting examples of failure cases:
Over very long rollouts, solids may become deformed
Some model seeds learn to predict large pieces of goop sticking to the wall instead of sliding down
See supplementary material for additional discussion.
Page updated
Google Sites
Report abuse