DiffCloth

Code repository for our paper DiffCloth: Differentiable Cloth Simulation with Dry Frictional Contact

📃 Paper | 🌍 Project

Tested Operating Systems

Ubuntu 22.04 | Mac OS 12

1. Download the repo:

Make sure to use the --recursive option to install the dependencies

git clone --recursive https://github.com/omegaiota/DiffCloth.git

2. Build CPP code with Cmake:

From the top directory:

mkdir build
cd build
cmake ..
make

3. Optimize/Visualize Section 6 Experiments:

• Run optimization:

  ./DiffCloth -demo {demooptions} -mode optimize -seed {randseed}

  where {demooptions} is the name of the demos from the following options and {randseed} is an integer for random initialization of the initial guesses of the tasks.

  The corresponding option for each of the experiments is:

  • 6.1 T-shirt: tshirt
  • 6.1 Sphere: sphere
  • 6.2 Hat: hat
  • 6.2 Sock: sock
  • 6.3 Dress: dress

• Visualize optimization iters:

  ./DiffCloth -demo {demooptions} -mode visualize -exp {expName}

  where {expName} is the iteration folder for visualization. The code repo comes with an example optimization run of T-shirt in output/tshirt-exampleopt/, and you can visualize the first iteration with

  ./DiffCloth -demo tshirt -mode visualize -exp tshirt-exampleopt/iter0/

The progress of the optimization is saved into the output/ directory of the root folder. Intermediate progress are visualized using the custom written OpenGL viewer.

4. Build Python Binding and Run Hat Controller example:

Build Python Binding:

• Install anaconda for virtual environment.
• In project root folder, run python setup.py install to install the python binding package. Rerun this command if you modify the CPP code.
• Create conda virtual environment: conda env create python=3.8 --file environment.yml, and activate it through conda activate diffcloth

Train/Test Hat Controller example:

• Navigate to src/python_code
• Test pretrained network: run python hatController.py --eval --render --load_expname 20210809-trainedBest
• Train network: run python hatController.py --render
• Resume train: run python hatController.py --train_resume --load_expname [expName] --load_epoch [epochNum]

Simulations are saved to the output/ directory of the root folder.

Note

Feel free to contact me at liyifei@csail.mit.edu or create a Github issue if you have questions regarding setting up the repository, running examples or adding new examples.

Citation

Please consider citing our paper if your find our research or this codebase helpful:

@article{Li2022diffcloth,
author = {Li, Yifei and Du, Tao and Wu, Kui and Xu, Jie and Matusik, Wojciech},
title = {DiffCloth: Differentiable Cloth Simulation with Dry Frictional Contact},
year = {2022},
issue_date = {February 2023},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
volume = {42},
number = {1},
issn = {0730-0301},
url = {https://doi.org/10.1145/3527660},
doi = {10.1145/3527660},
abstract = {Cloth simulation has wide applications in computer animation, garment design, and robot-assisted dressing. This work presents a differentiable cloth simulator whose additional gradient information facilitates cloth-related applications. Our differentiable simulator extends a state-of-the-art cloth simulator based on Projective Dynamics (PD) and with dry frictional contact [Ly et al. 2020]. We draw inspiration from previous work [Du et al. 2021] to propose a fast and novel method for deriving gradients in PD-based cloth simulation with dry frictional contact. Furthermore, we conduct a comprehensive analysis and evaluation of the usefulness of gradients in contact-rich cloth simulation. Finally, we demonstrate the efficacy of our simulator in a number of downstream applications, including system identification, trajectory optimization for assisted dressing, closed-loop control, inverse design, and real-to-sim transfer. We observe a substantial speedup obtained from using our gradient information in solving most of these applications.},
journal = {ACM Trans. Graph.},
month = {oct},
articleno = {2},
numpages = {20},
keywords = {differentiable simulation, cloth simulation, Projective Dynamics}
}

Counters (starts trakcing May 1 2024)