Stanford-ILIAD / Diverse-Conventions

Exploring techniques to generate diverse conventions in multi-agent settings

Home Page:http://iliad.stanford.edu/Diverse-Conventions/

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Diverse-Conventions

Source code for the NeurIPS 2023 paper "Diverse Conventions for Human-AI Collaboration."

To reproduce the results for the toy experiments (Blind Bandits and Balance Beam), use the "toy" branch. To reproduce the results for the Overcooked experiments, use the "neurips_comedi" branch.

For the Overcooked experiments, you can download pretrained weights and logs at this link. You can also download the raw (anonymized) user study data at this link.

About

Exploring techniques to generate diverse conventions in multi-agent settings

http://iliad.stanford.edu/Diverse-Conventions/

License:GNU General Public License v3.0


Languages

Language:JavaScript 96.4%Language:Python 2.8%Language:C++ 0.6%Language:HTML 0.1%Language:Shell 0.0%Language:CMake 0.0%Language:CSS 0.0%