COMPASS

What Went Wrong? Closing the Sim-to-Real Gap via Differentiable Causal Discovery

Conference on Robot Learning (CoRL), 2023

Please raise an issue or reach out to the authors if you need help with running the code.

Prepare Conda Env

Note: Our simulation environments are modified from robosuite.

conda env create -f environment.yml
pip install -e .

How to run

cd ./scripts/
conda activate compass

## You could start with our first air hockey environment and random action
python3 main.py --exp_name pusher

## Running code for drop environment and random action
python3 main.py --exp_name drop

## Running code for air hockey environment with SAC/PPO Agent training in the loop
python3 main.py --use_sac_agent 
python3 main.py --use_ppo_agent

There are a few parameters that you can tune from arguments, please feel free to modify them and discovery how COMPASS performs.

Logs will be saved using tensorboard

## Both the causal discovery and agent training will be saved under logdir folder
tensorboard --logdir= /Sim2Real-Compass/scripts/logdir/

Acknowledgement

Robosuite: A Modular Simulation Framework and Benchmark for Robot Learning
Stable-Baselines3

Citation

@misc{huang2023went,
      title={What Went Wrong? Closing the Sim-to-Real Gap via Differentiable Causal Discovery}, 
      author={Peide Huang and Xilun Zhang and Ziang Cao and Shiqi Liu and Mengdi Xu and Wenhao Ding and Jonathan Francis and Bingqing Chen and Ding Zhao},
      year={2023},
      eprint={2306.15864},
      archivePrefix={arXiv},
      primaryClass={cs.RO}
}