Supported Environments

The following environments are supported:

Control

Installation

pip install git+https://github.com/facebookresearch/mtenv.git@main#egg="mtenv[control]"

HiPBMDP

[ZSKP20] create a family of MDPs using the existing environment-task pairs from DeepMind Control Suite [TTM+20] and change one environment parameter to sample different MDPs. For more details, refer [ZSKP20].

Installation

pip install git+https://github.com/facebookresearch/mtenv.git@main#egg="mtenv[hipbmdp]"

Usage

from mtenv import make
env = make("MT-HiPBMDP-Finger-Spin-vary-size-v0")
env.reset()

MetaWorld

[YQH+20] proposed an open-source simulated benchmark for meta-reinforcement learning and multi-task learning consisting of 50 distinct robotic manipulation tasks. For more details, refer [YQH+20]. MTEnv provides a wrapper for the multi-task learning environments.

Installation

pip install git+https://github.com/facebookresearch/mtenv.git@main#egg="mtenv[metaworld]"

Usage

from mtenv import make
env = make("MT-MetaWorld-MT10-v0") # or MT-MetaWorld-MT50-v0 or MT-MetaWorld-MT1-v0
env.reset()

MPTE

Installation

pip install git+https://github.com/facebookresearch/mtenv.git@main#egg="mtenv[mpte]"

References

TTM+20

Yuval Tassa, Saran Tunyasuvunakool, Alistair Muldal, Yotam Doron, Siqi Liu, Steven Bohez, Josh Merel, Tom Erez, Timothy Lillicrap, and Nicolas Heess. Dm_control: software and tasks for continuous control. 2020. arXiv:2006.12983.

YQH+20(1,2)

Tianhe Yu, Deirdre Quillen, Zhanpeng He, Ryan Julian, Karol Hausman, Chelsea Finn, and Sergey Levine. Meta-world: a benchmark and evaluation for multi-task and meta reinforcement learning. In Conference on Robot Learning, 1094–1100. PMLR, 2020.

ZSKP20(1,2)

Amy Zhang, Shagun Sodhani, Khimya Khetarpal, and Joelle Pineau. Multi-task reinforcement learning as a hidden-parameter block mdp. arXiv preprint arXiv:2007.07206, 2020.