Peer Learning: Learning Complex Policies in Groups from Scratch via Action Recommendations

Multiple agents (peers) learning together simultaneously from scratch with the ability to communicate in a 'What would you do in my situation?' manner.

Installation

All packages are specified for Python version 3.7.4.

pip install -r requirements.txt
pip install stable-baselines3==1.5.0 --no-dependencies

Sample Usage

The entry point to Peer Learning is the run_peer.py file. You can simply do Peer Learning with the default setting by running the following line of code:

python run_peer.py

Experiments

In Experiments.md, you find the commands to replicate the experiments reported in our paper including all hyperparameters.

running pip install -r requirements_mogon.txt will result in stable-baselines3 1.5.0 requires gym==0.21, but you'll have gym 0.25.2 which is incompatible. but should work anyway.