Skip to content
/ dmfd Public

Learning Deformable Object Manipulation from Expert Demonstrations (RA-L, IROS 2022)

Notifications You must be signed in to change notification settings

uscresl/dmfd

Repository files navigation

Learning Deformable Object Manipulation from Expert Demonstrations

[Project website] [arXiv PDF] [IEEE manuscript]

This project is a PyTorch implementation of Learning Deformable Object Manipulation from Expert Demonstrations, published in IEEE RA-L and IEEE IROS, 2022.

SoftGym Installation

  1. Create new conda environment
conda create -n softgym python=3.7
  1. Install necessary packages for DMfD
cd dmfd
mkdir data # for storing experiments
conda env update -n softgym --file environment.yml
  1. Follow the docker guide for PyFlex installation

  2. Verify SoftGym is installed properly

# First, exit docker
conda activate softgym
export PYFLEXROOT=${PWD}/PyFlex
export PYTHONPATH=${PYFLEXROOT}/bindings/build:$PYTHONPATH
export LD_LIBRARY_PATH=${PYFLEXROOT}/external/SDL2-2.0.4/lib/x64:$LD_LIBRARY_PATH

4.1 (OPTIONAL) If you want these env variables set permanently, add these to your .bashrc

cd path_to_softagent/softgym
echo "export PYFLEXROOT=${PWD}/PyFlex" >> ~/.bashrc
echo "export PYTHONPATH=${PYFLEXROOT}/bindings/build:$PYTHONPATH" >> ~/.bashrc
echo "export LD_LIBRARY_PATH=${PYFLEXROOT}/external/SDL2-2.0.4/lib/x64:$LD_LIBRARY_PATH" >> ~/.bashrc

Run experiment

python examples/random_env.py --env_name PourWaterAmount # this should open up a GUI window running the PourWater task

SoftAgent Installation

  1. Update conda env with additional packages required by SoftAgent
# make sure you are at top level folder (/dmfd)
conda env update  --file environment.yml  --prune
  1. Activate the conda environment by running the following in terminal (outside of docker):
. ./prepare_1.0.sh

Note that prepare_1.0.sh only adds env variables to your current bash. Consider adding these variables to your .bashrc, for use on all future bash sessions.

Downloads

  • Download cached_initial_states and place it inside dmfd/softgym/softgym/ folder

  • For reproducing experimental results or training the behavior cloning agent, download checkpoints and place it on the top level of this project (dmfd/)

  • For training, DMfD and other Learning from Demonstration methods (BC, SAC-LfD, SAC-BC, AWAC) require expert demonstrations. See 'Generate Expert Demonstrations' below.

Running benchmarked experiments

  • To reproduce results of DMfD and some baselines (SAC, AWAC, BC, SAC-CURL, SAC-DrQ), checkout to the old branch to execute training/evaluation scripts. To reproduce results of other baselines (SAC-LfD and SAC-BC), checkout to the main branch to execute training/evaluation scripts. Make sure to recompile Softgym/PyFlex when you checkout a new branch (see docker guide).

  • To use the latest code (without the need to reproduce plots), feel free to use the code on the main branch for all experiments. Remember that you will need to recompile when switching to a new branch, and also regenerate demonstrations for the main branch.

Straighten Rope

############ Training ############

# State-based experiments:
sh ./scripts/train/state-dmfd-straighten-rope.sh # DMfD (Ours)
sh ./scripts/train/state-sac.sh # SAC - Set env=RopeFlatten
sh ./scripts/train/state-awac-straighten-rope.sh # AWAC
sh ./scripts/train/state-sac-lfd-straighten-rope.sh # SAC-LfD
sh ./scripts/train/state-bc-straighten-rope.sh # BC-state

# Image-based experiments:
sh ./scripts/train/image-dmfd-straighten-rope.sh # DMfD (Ours)
sh ./scripts/train/image-sac-curl.sh # SAC-CURL - Set env=RopeFlatten and env_kwargs_num_variations=1000
sh ./scripts/train/image-sac-drq.sh # SAC-DrQ - Set env=RopeFlatten and env_kwargs_num_variations=1000
sh ./scripts/train/image-sac-lfd-straighten-rope.sh # SAC-LfD
sh ./scripts/train/image-sac-bc-straighten-rope.sh # SAC-BC
sh ./scripts/train/image-bc-straighten-rope.sh # BC-image

############ Evaluation ############

# State-based experiments:
sh ./scripts/eval/state-dmfd-straighten-rope.sh # DMfD (Ours)
sh ./scripts/eval/state-sac-straighten-rope.sh # SAC
sh ./scripts/eval/state-awac-straighten-rope.sh # AWAC
sh ./scripts/eval/state-sac-lfd-straighten-rope.sh # SAC-LfD
sh ./scripts/eval/state-bc-straighten-rope.sh # BC-state

# Image-based experiments:
sh ./scripts/eval/image-dmfd-straighten-rope.sh # DMfD (Ours)
sh ./scripts/eval/image-sac-curl-straighten-rope.sh # SAC-CURL
sh ./scripts/eval/image-sac-drq-straighten-rope.sh # SAC-DrQ
sh ./scripts/eval/image-sac-lfd-straighten-rope.sh # SAC-LfD
sh ./scripts/eval/image-sac-bc-straighten-rope.sh # SAC-BC
sh ./scripts/eval/image-bc-straighten-rope.sh # BC-image

ClothFold

############ Training ############

# State-based experiments:
sh ./scripts/train/state-dmfd-clothfold.sh # DMfD (Ours)
sh ./scripts/train/state-sac.sh # SAC - Set env=ClothFold
sh ./scripts/train/state-awac-clothfold.sh # AWAC
sh ./scripts/train/state-sac-lfd-clothfold.sh # SAC-LfD
sh ./scripts/train/state-bc-clothfold.sh # BC-state

# Image-based experiments:
sh ./scripts/train/image-dmfd-clothfold.sh # DMfD (Ours)
sh ./scripts/train/image-sac-curl.sh # SAC-CURL - Set env=ClothFold and env_kwargs_num_variations=1000
sh ./scripts/train/image-sac-drq.sh # SAC-DrQ - Set env=ClothFold and env_kwargs_num_variations=1000
sh ./scripts/train/image-sac-lfd-clothfold.sh # SAC-LfD
sh ./scripts/train/image-sac-bc-clothfold.sh # SAC-BC
sh ./scripts/train/image-bc-clothfold.sh # BC-image

############ Evaluation ############

# State-based experiments:
sh ./scripts/eval/state-dmfd-clothfold.sh # DMfD (Ours)
sh ./scripts/eval/state-sac-clothfold.sh # SAC
sh ./scripts/eval/state-awac-clothfold.sh # AWAC
sh ./scripts/eval/state-sac-lfd-clothfold.sh # SAC-LfD
sh ./scripts/eval/state-bc-clothfold.sh # BC-state

# Image-based experiments:
sh ./scripts/eval/image-dmfd-clothfold.sh # DMfD (Ours)
sh ./scripts/eval/image-sac-curl-clothfold.sh # SAC-CURL
sh ./scripts/eval/image-sac-drq-clothfold.sh # SAC-DrQ
sh ./scripts/eval/image-sac-lfd-clothfold.sh # SAC-LfD
sh ./scripts/eval/image-sac-bc-clothfold.sh # SAC-BC
sh ./scripts/eval/image-bc-clothfold.sh # BC-image

ClothFoldDiagonalPinned

############ Training ############

sh ./scripts/train/image-dmfd-clothfolddiagonalpinned.sh # DMfD (Ours)
sh ./scripts/train/image-sac-curl.sh # SAC-CURL - Set env=ClothFoldRobot and env_kwargs_num_variations=1
sh ./scripts/train/image-sac-drq.sh # SAC-DrQ - Set env=ClothFoldRobot and env_kwargs_num_variations=1
sh ./scripts/train/image-sac-lfd-clothfolddiagonalpinned.sh # SAC-LfD
sh ./scripts/train/image-sac-bc-clothfolddiagonalpinned.sh # SAC-BC
sh ./scripts/train/image-bc-clothfolddiagonalpinned.sh # BC-image

############ Evaluation ############

sh ./scripts/eval/image-dmfd-clothfolddiagonalpinned.sh # DMfD (Ours)
sh ./scripts/eval/image-sac-curl-clothfolddiagonalpinned.sh # SAC-CURL
sh ./scripts/eval/image-sac-drq-clothfolddiagonalpinned.sh # SAC-DrQ
sh ./scripts/eval/image-sac-lfd-clothfolddiagonalpinned.sh # SAC-LfD
sh ./scripts/eval/image-sac-bc-clothfolddiagonalpinned.sh # SAC-BC
sh ./scripts/eval/image-bc-clothfolddiagonalpinned.sh # BC-image

ClothFoldDiagonalUnpinned

############ Training ############

sh ./scripts/train/image-dmfd-clothfolddiagonalunpinned.sh # DMfD (Ours)
sh ./scripts/train/image-sac-curl.sh # SAC-CURL - Set env=ClothFoldRobotHard and env_kwargs_num_variations=1
sh ./scripts/train/image-sac-drq.sh # SAC-DrQ - Set env=ClothFoldRobotHard and env_kwargs_num_variations=1
sh ./scripts/train/image-sac-lfd-clothfolddiagonalunpinned.sh # SAC-LfD
sh ./scripts/train/image-sac-bc-clothfolddiagonalunpinned.sh # SAC-BC
sh ./scripts/train/image-bc-clothfolddiagonalunpinned.sh # BC-image

############ Evaluation ############

sh ./scripts/eval/image-dmfd-clothfolddiagonalunpinned.sh # DMfD (Ours)
sh ./scripts/eval/image-sac-curl-clothfolddiagonalunpinned.sh # SAC-CURL
sh ./scripts/eval/image-sac-drq-clothfolddiagonalunpinned.sh # SAC-DrQ
sh ./scripts/eval/image-sac-lfd-clothfolddiagonalunpinned.sh # SAC-LfD
sh ./scripts/eval/image-sac-bc-clothfolddiagonalunpinned.sh # SAC-BC
sh ./scripts/eval/image-bc-clothfolddiagonalunpinned.sh # BC-image

Generate Expert Demonstrations (for DMfD and LfD baselines)

# Straighten Rope
 python experiments/generate_expert_trajs.py --save_observation_img=True --num_eps=8000 --env_img_size=32 --env_name=RopeFlatten --save_states_in_folder=False --out_filename=RopeFlatten_numvariations1000_eps8000_image_based_trajs.pkl

# ClothFold
 python experiments/generate_expert_trajs.py --save_observation_img=True --num_eps=8000 --env_img_size=32 --env_name=ClothFold --save_states_in_folder=True --out_filename=ClothFold_numvariations1000_eps8000_image_based_trajs.pkl

# ClothFoldDiagonalPinned
python experiments/generate_expert_trajs.py --save_observation_img=True --num_eps=1000 --env_img_size=32 --env_name=ClothFoldRobot --save_states_in_folder=True --num_variations=1 --out_filename=ClothFoldRobot_numvariations1_eps1000_image_based_trajs.pkl

# ClothFoldDiagonalUnPinned
python experiments/generate_expert_trajs.py --save_observation_img=True --num_eps=1000 --env_img_size=32 --env_name=ClothFoldRobotHard --save_states_in_folder=True --num_variations=1 --out_filename=ClothFoldRobotHard_numvariations1_eps1000_image_based_trajs.pkl

Cite

If you find this codebase useful in your research, please consider citing:

@article{salhotra2022dmfd,
	author={Salhotra, Gautam and Liu, I-Chun Arthur and Dominguez-Kuhne, Marcus and Sukhatme, Gaurav S.},
	journal={IEEE Robotics and Automation Letters},
	title={Learning Deformable Object Manipulation From Expert Demonstrations},
	year={2022},
	volume={7},
	number={4},
	pages={8775-8782},
	doi={10.1109/LRA.2022.3187843}
}

About

Learning Deformable Object Manipulation from Expert Demonstrations (RA-L, IROS 2022)

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published