SAC-GMM

Robot Skill Adaptation via Soft Actor-Critic Gaussian Mixture Models

Iman Nematollahi*, Erick Rosete Beas*, Adrian Röfer, Tim Welschehold, Abhinav Valada, Wolfram Burgard

We present SAC-GMM, ...

Installation

To begin, clone this repository locally

git clone https://github.com/nematoli/sac_gmm.git
export SACGMM_ROOT=$(pwd)/sac_gmm

Install requirements:

cd SACGMM_ROOT
conda create -n sacgmm_venv python=3.8
conda activate sacgmm_venv
sh install.sh

For Development:

pip install -r requirements-dev.txt
pre-commit install

Download

Download the CALVIN dataset and place it inside dataset/.

Skill Library

Step 1: Extract skill demos from the CALVIN dataset

Configure config/demos.yaml.

> python sacgmm/extract_demos.py skill='open_drawer'

Step 2: Train and evaluate skill libraries (Dynamical Systems) with ManifoldGMM

Configure config/gmm_train.yaml.

> python sac_gmm/scripts/gmm_train.py skill='open_drawer'

Configure config/gmm_eval.yaml.

> python sac_gmm/scripts/gmm_eval.py skill='open_drawer'

Pre-trained Models

We provide our final models for ...

cd SACGMM_ROOT/checkpoints
sh download_model_weights.sh

Training

python 

Ablations

python 

Evaluation

python 

Citation

If you find the code useful, please cite:

SAC-GMM

@inproceedings{nematollahi22icra,
    author  = {Iman Nematollahi and Erick Rosete-Beas and Adrian Roefer and Tim Welschehold and Abhinav Valada and Wolfram Burgard},
    title   = {Robot Skill Adaptation via Soft Actor-Critic Gaussian Mixture Models},
    booktitle = {Proceedings of the IEEE International Conference on Robotics and Automation  (ICRA)},
    pages={8651-8657},
    year = 2022,
    url={http://ais.informatik.uni-freiburg.de/publications/papers/nematollahi22icra.pdf},
    address = {Philadelphia, USA}
}

License

MIT License