All Tools
P
AgentsFreeOpen Source
PYMARL
Deep multi-agent RL framework behind SMAC algorithms
Apache-2.0
ABOUT
Cooperative MARL papers need a shared training loop for value factorization methods. PyMARL packages QMIX, COMA, VDN, IQL, and QTRAN on StarCraft SMAC maps so researchers compare cooperative agents without reimplementing each algorithm from scratch.
INTEGRATION GUIDE
1. Train QMIX or VDN agents on SMAC maps like 2s3z and 3s5z
2. Compare COMA counterfactual gradients with independent Q-learning
3. Save and reload MARL checkpoints for SMAC evaluation replays
4. Reproduce SMAC paper baselines with a shared PyTorch trainer
TAGS
pythonagentsmulti-agentreinforcement-learningmarlpytorchstarcraft