Implementation of multiple Multi-Agent Reinforcement Learning (MARL) methods on the overcooked environment using JAX. The benchmark includes two MARL algorithms, as well as 5 CL methods. COOX is the first benchmark designed for Continual Multi-Agent Reinforcement Learning. It is also the first CL benchmark to use JAX, significantly speeding up computation.
The benchmark includes 36 Overcooked layouts, of varying difficulty level. Here we show 3 examples of layouts.
First create a virtual environment
python3.10 -m venv .venv
Second, activate this virtual environment:
source .venv/bin/activate
.venv\Scripts\activate.bat
Then, install the requirements:
pip install -r requirements.txt
(for Mac: first remove the [cuda12] in requirements.txt)
Another option is to install with conda:
conda create -n jaxovercooked python=3.10
conda activate jaxovercooked
pip install -r requirements.txt
After installing, the IPPO implementation can be run from the base folder:
python -m baselines.IPPO_CL
python -m baselines.IPPO_MLP