SkillDiffuser

Zhixuan Liang, Yao Mu, Hengbo Ma, Masayoshi Tomizuka, Mingyu Ding, Ping Luo

Usage

Setup Python Environment

1. Install MuJoCo 200

unzip mujoco200_linux.zip
mv mujoco200_linux mujoco200
cp mjkey.txt ~/.mujoco
cp mjkey.txt ~/.mujoco/mujoco200/bin

# test the install
cd ~/.mujoco/mujoco200/bin
./simulate ../model/humanoid.xml

# add environment variables
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:~/.mujoco/mujoco200/bin
export MUJOCO_KEY_PATH=~/.mujoco/${MUJOCO_KEY_PATH}

2. Install Pypi Packages

pip install -r requirements.txt

3. Install LOReL Environment

git clone https://github.com/suraj-nair-1/lorel.git

cd lorel/env
pip install -e .

Setup LOReL Dataset

1. Download the dataset from LOReL
2. Process the dataset from h5py to pickle

python h5py2pkl.py --root_path <path to dir of may_08_sawyer_50k> --output_name <output_file_name.pkl>

3. Change the path in hrl/conf/env/lorel_sawyer_obs.yaml to the processed dataset.

Instructions

Our code for running SkillDiffuser experiments is present in hrl folder.

To run the code, please use the following command:

./train_lorel_compose.sh

This is a sample command intended to show the usage of different flags available. The checkpoints can be downloaded from here. (The checkpoint is used for fine-tuning, not for evaluation directly.)

If you would like to evaluate the model directly, please see this issue.

License

The code is made available for academic, non-commercial usage.

For any inquiry, contact: Zhixuan Liang ([email protected])