Codebase for the self-supervised goal reaching benchmark introduced in the LEXA paper (Discovering and Achieving Goals via World Models, NeurIPS 2021).

Create the conda environment by running : conda env create -f environment.yml

Alternatively, you can update an existing conda environment by running : conda env update -f environment.yml

Modify the python path
export PYTHONPATH=<path to lexa-benchmark>

Export the following variables for rendering
export MUJOCO_RENDERER=egl; export MUJOCO_GL=egl

Please follow these instructions to install mujoco

WARNING! The success criteria defined in this benchmark should not be used as a reward for the agent. The criteria were tuned to roughly match human intuition, but produce some false positives, which the agent can exploit if it observes the success variable.

If you find this code useful, please cite:

@misc{lexa2021,
    title={Discovering and Achieving Goals via World Models},
    author={Mendonca, Russell and Rybkin, Oleh and
    Daniilidis, Kostas and Hafner, Danijar and Pathak, Deepak},
    year={2021},
    Booktitle={NeurIPS}
}

This benchmark is built on top of the following environments: Adept, MetaWorld, and DeepMind Control Suite.