/PIC

PIC: Permutation Invariant Critic for Multi-Agent Deep Reinforcement Learning

Primary LanguagePythonOtherNOASSERTION

PIC: Permutation Invariant Critic for Multi-Agent Deep Reinforcement Learning

CORL 2019

[Project Website] [PDF]

Iou-Jen Liu*, Raymond A. Yeh*, Alexander G. Schwing
University of Illinois at Urbana-Champaign
(* indicates equal contribution)

The repository contains Pytorch implementation of MADDPG with Permutation Invariant Critic (PIC).

If you used this code for your experiments or found it helpful, please consider citing the following paper:

@inproceedings{LiuCORL2019,
  author = {I.-J. Liu$^\ast$ and R.~A. Yeh$^\ast$ and A.~G. Schwing},
  title = {PIC: Permutation Invariant Critic for Multi-Agent Deep Reinforcement Learning},
  booktitle = {Proc. CORL},
  year = {2019},
  note = {$^\ast$ equal contribution},
}

Platform and Dependencies:

Install the improved MPE:

cd multiagent-particle-envs
pip install -e .

Please ensure that multiagent-particle-envs has been added to your PYTHONPATH.

Training

cd maddpg
python main_vec.py --exp_name coop_navigation_n6 --scenario simple_spread_n6  --critic_type gcn_max  --cuda

Acknowledgement

The MADDPG code is based on the DDPG implementation of https://github.com/ikostrikov/pytorch-ddpg-naf

The improved MPE code is based on the MPE implementation of https://github.com/openai/multiagent-particle-envs

The GCN code is based on the implementation of https://github.com/tkipf/gcn

License

PIC is licensed under the MIT License