saptab/multi-agent-curriculum

Multi-Agent-Curriculum

The library's goal is to support current and evolving multi-agent algorithms with curriculums.

Current multi-agent algorithms supported from an updated EPyMARL library include on-policy algorithms like MAPPO, MAA2C, COMA and off-policy algorithms like MADDPG, QMIX, VDN.

The updates involve supporting curriculum features with multi-agent reinforcement learning algorithms to run either episodically or parallely. The code updates are in src/run.py, src/envs/__init__.py, src/runners/episode_runner.py, src/runners/parallel_runner.py

Domain Randomization feature is provided.

The environment supported for now is Level Based Foraging.

New algorithms, curriculums and environments are being developed.

Installation Instruction

Conda Environment with Python 3.7 can be created as follows

conda create -n marl_cc python=3.7
conda activate marl_cc

Then the dependencies can be installed with

cd marl_dr_curr_epymarl
pip install -r requirements.txt
pip install torch_scatter==2.1.1
pip install wandb
pip install protobuf==3.20.*

torch_scatter is installed separately as it requires torch to be already installed

Shell scripts under code_scripts folder can be used to run the Multi-Agent Reinforcement Learning algorithms with Domain Randomization codes on Level Based Foraging.