DiffclothAI for knotting tie

Dependencies

• libccd: https://github.com/danfis/libccd, follow the instruction to install libccd.
• fcl: https://github.com/flexible-collision-library/fcl, follow the instruction to install fcl. Perhaps you also need to install eigen3 manually. I use sudo apt install libeigen3-dev to install eigen3 library.
• Tight-Inclusion: https://github.com/Continuous-Collision-Detection/Tight-Inclusion, clone Tight-Inclusion in the root directory of this repository.
• yaml-cpp: https://github.com/jbeder/yaml-cpp, follow the instruction to install yaml-cpp

Basic Usage

1. Clone the repository with --recursive
2. Install dependencies.
3. Create a conda environment: conda create -n knotting_tie python=3.8. You can also create a conda environment follow the instruction in DiffCloth. It will create a python3.7 environment. But I found python3.8 also works well. Then, install torch, numpy, PyYAML, and other package you need.
4. Run python setup.py install at the root directory of this repository. It will automatically compile DiffClothAI and install python interface.
5. Start configuering your tasks. First you need to define the mesh parameters and scene configuration. These are written in a yaml file. You can take towel_config.yaml for example. Then, to write you own script, you can start with src/python_code/repair.py for how to load a mesh into DiffClothAI and drag some vertices on it to control the mesh. A more complex example is src/python_code/action_select.py. It will set an optimization objective and use gradient descent to optimize a_control to control the mesh.

real2sim related files: src/python_code/fitting_alt.py

learn@sim related files: src/python_code/RL_tie.py, action_select_env.py