YilingQiao/diff_fem

Differentiable Simulation of Soft Multi-body Systems (NeurIPS 2021)

Differentiable Simulation of Soft Multi-body Systems

Yi-Ling Qiao, Junbang Liang, Vladlen Koltun, Ming C. Lin

[Paper] [Code] [Video]

Updates

1. Link to a demo video.
2. cmakelist and setup files.
3. demos for inverse problems and forward simulation.

TODO

1. More demos for the control problems.
2. More documents for the config files.
3. Utils for make tet meshes.
4. More readme documentation.

Setup

1. Create a conda virtual environment and activate it.

conda create -n difem python=3.8 -y
conda activate difem

2. Download and build the project.

git clone git@github.com:YilingQiao/diff_fem.git
cd diff_fem
git submodule init
git submodule update
sudo apt-get install ninja-build cppad libcgal-dev
python setup.py install

3. Run the examples

Examples

Inverse problem

1. Suspension bridge (Fig. 3a in the paper)

python python/demo_sus.py

2. Arch bridge (Fig. 3c in the paper)

python python/demo_br.py

For the above 2 experiments, the output meshes are stored in out/

Control Problems

1. Drone (Fig. 4a in the paper) TODO

2. Octopus (Fig. 4b in the paper)

python python/demo_octopus.py

3. Fish (Fig. 4c in the paper) TODO

For the above experiments, the output meshes are stored in out_test/

Forward Simulation

Note that our simulator can be used for pure forward simulation. In this case, we replace the autodiff scalar (cppad) by C++ double and can run much faster (more than 5x).

To make this change, we first uncomment

// #define FORWARD_ONLY

in python/pydifem.cc and then run

python setup.py install

1. Baymax (Fig. 1 in the paper)

python python/demo_baymax.py

2. Clothball (Fig. 2 in the paper)

python python/demo_cloth_ball.py

For the above experiments, the output meshes are stored in out_test/

Our Related Repos

Differentiable Soft Body Dynamics (this repo) Code Paper Differentiable Simulation of Soft Multi-body Systems. Yi-Ling Qiao, Junbang Liang, Vladlen Koltun, Ming C. Lin. (Neurips 2021)

Differentiable Articulated Body Dynamics Code Paper Efficient Differentiable Simulation of Articulated Bodies. Yi-Ling Qiao, Junbang Liang, Vladlen Koltun, Ming C. Lin. (ICML 2021)

Differentiable Dynamics for Rigid Body and Cloth Coupling Code Paper Scalable Differentiable Physics for Learning and Control. Yi-Ling Qiao, Junbang Liang, Vladlen Koltun, Ming C. Lin. (ICML 2020)

Differentiable Cloth Dynamics Code Paper Differentiable Cloth Simulation for Inverse Problems. Junbang Liang, Ming C. Lin, Vladlen Koltun. (NeurIPS 2019)

Bibtex

@inproceedings{Qiao2021Differentiable,
author  = {Qiao, Yi-Ling and Liang, Junbang and Koltun, Vladlen and Lin, Ming C.},
title  = {Differentiable Simulation of Soft Multi-body Systems},
booktitle = {Conference on Neural Information Processing Systems (NeurIPS)},
year  = {2021},
}