MA-POCA Agent playing SoccerTwos

This repository contains a trained multi-agent reinforcement learning model for the Unity ML-Agents SoccerTwos environment.

Environment

• Environment: SoccerTwos
• Teams: 2
• Agents per team: 2
• Trainer: MA-POCA (POCA)
• Self-play: Enabled
• Framework: Unity ML-Agents
• Model format: ONNX
• Training steps: 547,682

Model

The trained policy is available as:

SoccerTwos.onnx

Training

The model was trained using the official Unity ML-Agents SoccerTwos environment and the MA-POCA trainer with self-play.

Training was stopped after 547,682 steps and the final ONNX model was exported successfully.

The reported reward metadata is based on training group-reward observations and is not an AI-vs-AI leaderboard score.

Hugging Face Deep Reinforcement Learning Course

This model was trained for Unit 7 of the Hugging Face Deep Reinforcement Learning Course.