ACM Transactions on Graphics (Proceedings of SIGGRAPH Asia), December 2023.
Quentin Becker · Seiichi Suzuki · Yingying Ren · Davide Pellis · Julian Panetta · Mark Pauly

This repository contains the source code and data for the paper C-shells: Deployable Gridshells with Curved Beams, published at SIGGRAPH Asia 2023.

A C-shell is composed of curved flexible beams connected at rotational joints that can be assembled in a stress-free planar configuration. When actuated, the elastic beams deform and the assembly deploys towards the target 3D shape.

Before cloning the repository, make sure you have installed all the dependencies required by elastic rods. The design optimization and planarization parts of the code also depend on the commercial optimization package knitro; these will be omitted from the build if knitro is not found.

Start by cloning the repository:

git clone --recursive git@github.com:EPFL-LGG/Cshells.git
cd Cshells

The dependencies can be installed as follow

conda create -n cshell_env python=3.8
conda activate cshell_env
conda install -y pytorch=2.0 -c pytorch
conda install -y scipy matplotlib
conda install -y 'jupyterlab>=3.2'
pip install geomdl==5.3

To deactivate the environment, simply execute:

conda deactivate

You can build the project as follow (you need to have the environment activated):

mkdir build && cd build
cmake .. -GNinja
ninja

Install pythreejs using

conda install -y pythreejs -c conda-forge
jupyter lab build

If this fails, you can try to install the source. Go to the ext/ folder and clone:

git clone https://github.com/jpanetta/pythreejs
cd pythreejs
pip3 install -e .
cd js
jupyter labextension install .

To run a notebook, first launch jupyter lab in the one of the parent folders:

jupyter lab

Make sure you have activated the environment first.

Three main actions can be performed on C-shells: deployment, design optimization, and planarization.

A CShell object (defined in python/CShell.py) can be instanciated and deployed as shown in the notebook notebooks/deployments/deployments.ipynb. A plurality of designs to deploy can be found under data/models. A linkage can be deployed either by specifying an attraction surface (see src/AverageAngleSurfaceAttractedLinkage.hh) or by fixing some degrees of freedom (see src/AverageAngleLinkage.hh). We show different deployment strategies in notebooks/deployments/deployments_torus_symmetric.ipynb

Once deployed, a CShell can be optimized towards a prescribed target surface while keeping the elastic energy in the deformed state low. Some examples can be found under notebooks/design_optimizations. In particular optim_torus_symmetric.ipynb shows an example of radially symmetric design. The notebook optim_hexagon_xshell.ipynb optimizes a design using the X-shell method in the current code framework.

If a target surface in the form of a B-spline surface is known in advance, one may apply the planarization algorithm as shown in notebooks/planarizations. The joints are laid out in a plane and optimized so that the C-shell formed by connecting the joints is a good initial guess for further design optimization towards the user defined target surface.