This repository contains the code for the composed shielding experiments with maddpg Safe multi-agent reinforcement learning via shielding paper.
- Python 3.6+
- gym
- matplotlib 3.0.0
- particle environment for deep MARL experiments (modified to be discretized + scenarios - code missing due to computer problems).
- shields for the particle environment synthesized using the Slugs tool, not provided because of file sizes.
GridShield.py: contains the implementation of the composed shielding method currently restricted to 2 agents per shield but code can be modified to accomodate more.train_maddpg.py: train for a given scenario and record information with shielding using composed shielding option.train_test.py: train then run testing phase (no learning or exploration) and record relevant information./logs: contains the output logged/policy: contains policy checkpoints for maddpg/learning_curves: contains relevant info for graphing (rewards and collisions)./benchmark_files: contains info pertaining to collisions for shielding and without shielding.
- Code is provided as is and not actively maintained at the moment. However, I am happy to answer questions.