Sparse Meta-World

This is a set of visual sparse-reward metaworld environments used in our ICML 2021 paper, LatCo. We support the following 7 environments.

This code is modified from the original MetaWorld for better non-exploitable sparse rewards and easier installation. Please refer to the LatCo paper for more details and below for installation and usage instructions.

Installation and Usage

Sparse Meta-World is based on MuJoCo, which has a proprietary dependency we can't set up for you. Please follow the instructions in the mujoco-py package for help. Once you're ready to install everything, run:

pip install git+https://github.com/zchuning/metaworld.git@master#egg=metaworld

Alternatively, you can clone the repository and install an editable version locally:

git clone https://github.com/zchuning/metaworld.git
cd metaworld
pip install -e .

Now you can simply import the environment from python:

from metaworld import SparseMetaWorld
import numpy as np
env = SparseMetaWorld('SawyerReachEnvV2', 2)
o = env.reset()
o,r,d,i = env.step(np.zeros(4))

The environment does not return a dense reward, only the 0/1 sparse reward.

By default, the rendering will use glfw. With egl rendering, gpu 0 will be used by default. You can change these with the following environment variables

export MUJOCO_RENDERER='egl'
export GL_DEVICE_ID=1

Bibtex

If you use the Sparse MetaWorld benchmark for academic research, please kindly cite

@inproceedings{rybkin2021latco,
  title={Model-Based Reinforcement Learning via Latent-Space Collocation},
  author={Rybkin, Oleh and Zhu, Chuning and Nagabandi, Anusha and Daniilidis, Kostas and Mordatch, Igor and Levine, Sergey},
  journal={Proceedings of the 38th International Conference on Machine Learning},
  year={2021}
}

@inproceedings{yu2019meta,
  title={Meta-World: A Benchmark and Evaluation for Multi-Task and Meta Reinforcement Learning},
  author={Tianhe Yu and Deirdre Quillen and Zhanpeng He and Ryan Julian and Karol Hausman and Chelsea Finn and Sergey Levine},
  booktitle={Conference on Robot Learning (CoRL)},
  year={2019}
  eprint={1910.10897},
  archivePrefix={arXiv},
  primaryClass={cs.LG}
  url={https://arxiv.org/abs/1910.10897}
}

Acknowledgements

The original MetaWorld repo is at https://github.com/rlworkgroup/metaworld.