thomasweng15/FabricFlowNet

Code for the CoRL 2021 paper "FabricFlowNet: Bimanual Manipulation using a Flow-based Policy"

FabricFlowNet: Bimanual Cloth Manipulation with a Flow-based Policy

Thomas Weng, Sujay Bajracharya, Yufei Wang, Khush Agrawal, David Held Conference on Robot Learning 2021

[arXiv] [pdf] [project page]

This repo contains the code for running the FabricFlowNet model used to report the simulation results in the paper.

The ROS code for running FabricFlowNet on a real-world system will be released in a separate repository.

File Structure

├── FabricFlowNet/
|   |── data/                   # Folder for datasets, saved runs, evaluation goals 
|   └── fabricflownet/
|       |── flownet/            # FlowNet model and training code
|       |── picknet/            # PickNet model and training code
|       |── eval.py             # Evaluation script
|       └── utils.py            
└── softgym/                    # SoftGym submodule

Installation

These instructions have been tested on Ubuntu 18.04 with NVIDIA GTX 3080/3090 GPUs. SoftGym requires CUDA 9.2+, FFN training and inference have been tested on CUDA 11.1.

• Clone this repo
• Set up the softgym submodule, tracking the fabricflownet branch: git submodule update --init
• Follow the softgym instructions to create a softygm conda environment. Then compile PyFlex: . prepare_1.0.sh and . compile_1.0.sh. Check the compile script to make sure that the CUDA_BIN_PATH env variable is set to the path of the CUDA library you installed SoftGym with.
• In the FabricFlowNet directory, activate the conda environment and set environment variables: . prepare_1.0.sh
• conda env update -f environment.yml --prune. Ensure that you install a PyTorch version that is suitable for your CUDA version.
• Install the repo as a python package: pip install -e .

Evaluation

• Download the evaluation set and model weights into ./data/:
  • Download and extract the evaluation set
  • Download the FlowNet weights, this does not need to be extracted
  • Download and extract the PickNet weights
• Run the evaluation script: python fabricflownet/eval.py --run_path=data/picknet_run --ckpt=105000 --cloth_type=square_towel
  • To run in headless mode, add the --headless flag; use the -h flag to see other available flags.
• Performance on square towel goals from the paper (in mm):

Square Towel (mm)
all: 6.803 +/- 10.413
one-step: 4.262 +/- 2.287
mul-step: 23.741 +/- 21.597

Rectangular Towel (mm)
all: 9.270 +/- 7.001
one-step: 4.254 +/- 1.086
mul-step: 16.793 +/- 5.143

Tshirt (mm)
all: 31.628 +/- 10.892
one-step: 24.459 +/- 4.876
mul-step: 45.965 +/- 0.000

Citation

If you find this code useful in your research, please feel free to cite:

@inproceedings{weng2021fabricflownet,
 title={FabricFlowNet: Bimanual Cloth Manipulation with a Flow-based Policy},
 author={Weng, Thomas and Bajracharya, Sujay and Wang, Yufei and Agrawal, Khush and Held, David},
 booktitle={Conference on Robot Learning},
 year={2021}
}