All Tools
J
AgentsFreeOpen Source
JAXMARL
Multi-agent RL environments and baselines in JAX
Apache-2.0
ABOUT
MARL experiments stall when each environment needs its own CPU simulator and baselines live in separate codebases. JaxMARL vectorizes popular multi-agent tasks in JAX, including SMAX as a StarCraft-style benchmark without the game engine, so algorithms can train and compare on GPU.
INSTALL
pip install jaxmarlINTEGRATION GUIDE
1. Train MAPPO or IPPO baselines on SMAX, MPE, Hanabi, or Overcooked in JAX
2. Benchmark cooperative MARL without installing the StarCraft II engine
3. Vectorize multi-agent Brax robotics control on GPU for faster sweeps
4. Compare new MARL algorithms against JaxMARL's maintained reference implementations
TAGS
pythonjaxmarlmulti-agentreinforcement-learninggpu