An implementation of maddpg-pytorch on a multi-agent minigrid environment.

• OpenAI baselines, commit hash: 98257ef8c9bd23a24a330731ae54ed086d9ce4a7
• shariqiqbal2810's fork of Multi-agent Particle Environments
• PyTorch, version: 1.6.0
• OpenAI Gym, version: 0.9.6
• Tensorboard, version: 2.3.0 and tensorboardX, version: 2.1 (for logging)
• My fork of MiniGrid

The versions are just what I used and not necessarily strict requirements.

All training code is contained within main.py. To view options simply run:

python main.py --help

To train an example in the MiniGrid env:

ma-minigrid-maddpg: python main.py --env_id MiniGrid-MA-UnlockDoorGoalA2-v0 --model_name test_minigrid --discrete_action

To visualize a model from the MiniGrid env:

python evaluate.py --env_id MiniGrid-MA-UnlockDoorGoalA2-v0 --model_name test_minigrid

To train an example in the MPE env:

python main.py --mpe

To visualize a model from the MPE env:

python evaluate.py --mpe

maddpg-pytorch & all acknowledgements from that repo as well.