Multi HRI

Set up guide

1. Follow setup instrcution for overcooked_ai from here
2. Clone this repository: git clone [email protected]:HIRO-group/multiHRI.git
3. Activate conda env: conda activate mHRI
4. Run: pip install pip==24.0 wheel==0.38.4 setuptools==65.5.0
5. cd into the repo and run: pip install -e ..
6. Install liblsl using conda

Training agents

Refer to scripts/train_agents.py for examples on how to train different agent trainings. If you've installed the package as above, you can run the script using:
python scripts/train_agents.py

Citation

If you use this repository in any way, please cite:

{@inproceedings{10.5555/3545946.3598926,
author = {Aroca-Ouellette, St\'{e}phane and Aroca-Ouellette, Miguel and Biswas, Upasana and Kann, Katharina and Roncone, Alessandro},
title = {Hierarchical Reinforcement Learning for Ad Hoc Teaming},
year = {2023},
isbn = {9781450394321},
publisher = {International Foundation for Autonomous Agents and Multiagent Systems},
address = {Richland, SC},
abstract = {When performing collaborative tasks with new unknown teammates, humans are particularly adept at adapting to their collaborator and converging toward an aligned strategy. However, state of the art autonomous agents still do not have this capability. We propose that a critical reason for this disconnect is that there is an inherent hierarchical structure to human behavior that current agents lack. In this paper, we explore the use of hierarchical reinforcement learning to train an agent that can navigate the complexities of ad hoc teaming at the same level of abstraction as humans. Our results demonstrate that when paired with humans, our Hierarchical Ad Hoc Agent (HAHA) outperforms all baselines on both the team's objective performance and the human's perception of the agent.},
booktitle = {Proceedings of the 2023 International Conference on Autonomous Agents and Multiagent Systems},
pages = {2337–2339},
numpages = {3},
keywords = {ad hoc teaming, hierarchical reinforcement learning, human agent collaboration, mutual adaptation, zero-shot coordination},
location = {London, United Kingdom},
series = {AAMAS '23}
}