python3 diffmpm.py

DiffTaichi: Differentiable Programming for Physical Simulation (ICLR 2020)

Yuanming Hu, Luke Anderson, Tzu-Mao Li, Qi Sun, Nathan Carr, Jonathan Ragan-Kelley, Frédo Durand

[Paper] [Video] (with instructions to reproduce every demo)

Differentiable programming in Taichi allows you to optimize neural network controllers efficiently with brute-force gradient descent, instead of using reinforcement learning.

The DiffTaichi differentiable programming framework is now officially part of Taichi. This repo only contains examples.

DiffTaichi significantly boosts the performance and productivity of differentiable physical simulators. For example, the differentiable elastic object simulator (ChainQueen) in DiffTaichi is 188x faster than an implementation in TensorFlow. The DiffTaichi version also runs as fast as the CUDA implementation, with the code being 4.2x shorter.

Most of the 10 differentiable simulators can be implemented within 2-3 hours.

Questions regarding the simulators/autodiff compiler go to Yuanming Hu (yuanming at mit.edu) or Issues.

Note: Updates on October 27 2021

Examples are now compatible with Taichi v0.8.3. Please update Taichi if you are using an old version.

How to run

Step 1: Install Taichi with pip:

(Most examples do not need a GPU to run.)

python3 -m pip install taichi

Step 2: Run example scripts in the examples folder: (Please wait for all GIFs to load :-)

Differentiable Elastic Object Simulator [python3 diffmpm.py]

Gradient descent iteration 0 and gradient descent iteration 80:

Differentiable 3D Elastic Object Simulator [python3 diffmpm3d.py]

Gradient descent iteration 40:

Differentiable 3D Fluid Simulator [python3 liquid.py]

Gradient descent iteration 450:

Differentiable Height Field Water Simulator [python3 wave.py]

Gradient descent iteration 180:

Differentiable (Adversarial) Water Renderer [python3 water_renderer.py]

Differentiable water simulation + differentiable water rendering + (differentiable) CNN

Optimization goal: find an initial water height field, so that after simulation and shading, VGG16 thinks the squirrel image is a goldfish. Input image: VGG16=fox squirrel (42.21%)

Left: center activation . Right: An activation that fools VGG (VGG16=goldfish (99.91%))

Differentiable Rigid Body Simulator [python3 rigid_body.py [1/2] train]

2048 time steps. Gardient descent iteration 20:

Differentiable Mass-Spring Simulator [python3 mass_spring.py [1/2/3] train]

682 time steps. Gardient descent iteration 20:

Differentiable Billiard Simulator [python3 billiards.py]

Gardient descent iteration 0 and gradient descent iteration 100:

See the video for the remaining two simulators.

Bibtex

@article{hu2019difftaichi,
  title={DiffTaichi: Differentiable Programming for Physical Simulation},
  author={Hu, Yuanming and Anderson, Luke and Li, Tzu-Mao and Sun, Qi and Carr, Nathan and Ragan-Kelley, Jonathan and Durand, Fr{\'e}do},
  journal={ICLR},
  year={2020}
}
@article{hu2019taichi,
  title={Taichi: a language for high-performance computation on spatially sparse data structures},
  author={Hu, Yuanming and Li, Tzu-Mao and Anderson, Luke and Ragan-Kelley, Jonathan and Durand, Fr{\'e}do},
  journal={ACM Transactions on Graphics (TOG)},
  volume={38},
  number={6},
  pages={201},
  year={2019},
  publisher={ACM}
}