This is a trained multi-agent team playing SoccerTwos using the Unity ML-Agents Library with MA-POCA (Multi-Agent POsthumous Credit Assignment) and self-play.
The agents learned cooperative 2v2 soccer behavior where each team tries to score goals while preventing the opponent from scoring.
Algorithm: MA-POCA with Self-Play Training: 5,000,000 timesteps Environment: SoccerTwos 2v2 soccer Team Strategy: Cooperative multi-agent behavior
This model participates in the AI vs AI Challenge where it competes against other trained agents on the leaderboard.
