Introdution

This repo is the code for human-AI experiments on Overcooked by AAMAS 2023 paper PECAN.

For code of PECAN, please refer to this repo.

We integrate Human-Aware-RL agent models with the PantheonRL framework for convenient human-ai coordination study on Overcooked. Changes are done under the overcookedgym/overcooked-flask directory.

Instruction for usage

1. Install libraries

Install PantheonRL in this repo

   pip install -e .

Install Overcooked in this repo

   pip install -e overcookedgym/human_aware_rl/overcooked_ai

2. Save models

Save models of Human-Aware-RL agents in this directory (like the given MEP model for layout simple)

3. Start a process

For example, this will start a process on port 8008 with an MEP agent on the layout simple.

   python overcookedgym/overcooked-flask/app.py --layout=simple --algo=0 --port=8008 --seed=1

The next command will start a dummy demo agent.

   python overcookedgym/overcooked-flask/app.py --layout=simple --port=8008 --dummy=True

Citation

Please cite

@article{lou2023pecan,
 title={PECAN: Leveraging Policy Ensemble for Context-Aware Zero-Shot Human-AI Coordination},
 author={Lou, Xingzhou and Guo, Jiaxian and Zhang, Junge and Wang, Jun and Huang, Kaiqi and Du, Yali},
 journal={arXiv preprint arXiv:2301.06387},
 year={2023}
}

@inproceedings{sarkar2022pantheonrl,
 title={PantheonRL: A MARL Library for Dynamic Training Interactions},
 author={Sarkar, Bidipta and Talati, Aditi and Shih, Andy and Sadigh, Dorsa},
 booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
 volume={36},
 number={11},
 pages={13221--13223},
 year={2022}
}

@article{carroll2019utility,
 title={On the utility of learning about humans for human-ai coordination},
 author={Carroll, Micah and Shah, Rohin and Ho, Mark K and Griffiths, Tom and Seshia, Sanjit and Abbeel, Pieter and Dragan, Anca},
 journal={Advances in neural information processing systems},
 volume={32},
 year={2019}
}