Causal Curiosity
Official implementation of Causal Curiosity: RL Agents Discovering Self-supervised Experiments for Causal Representation Learning at ICML 2021. Paper and Website
Installation
Download our version of CausalWorld from this drive link. Once downloaded add it to this repository and follow instructions to install CausalWorld.
You will also need mujoco-py. Follow the installation instructions here.
After installing mujoco-py, you will need to edit the done property for each of the mujoco agents property files. The done
property needs to be set to False
. Otherwise the environment will stop simulating if the agents orientation exceeds a threshold.
Usage
For Mujoco experiments, run
python pnw_mujoco.py
For CausalWorld experiments, run
python plan_and_write_video_vanilla_cw.py
Citation
@article{sontakke2020causal,
title={Causal Curiosity: RL Agents Discovering Self-supervised Experiments for Causal Representation Learning},
author={Sontakke, Sumedh A and Mehrjou, Arash and Itti, Laurent and Sch{\"o}lkopf, Bernhard},
journal={arXiv preprint arXiv:2010.03110},
year={2020}
}