MAgent

Why we included this project

MAgent is a research platform for many-agent reinforcement learning, built to take experiments beyond single- or few-agent setups and scale up to hundreds or millions of agents acting at once. The included tasks, such as pursuit and gathering, plus a battle mode, give you ready-made benchmarks, and the Python interface lets you pair hand-written rule-based agents with deep-learning policies. Since the original repo is no longer maintained, treat it as a historical reference and use the actively developed MAgent2 fork from the Farama Foundation for new installs. If you are reproducing published results or want to compare your own many-agent methods against established tasks, this is the environment those numbers came from.