A Unity-based Air Hockey simulation leveraging ML-Agents for reinforcement learning agent training and inference.

Multi-agent reinforcement learning in Unity’s Soccer Twos environment using POCA. Features enhanced observation memory, custom reward shaping, and optimized training configurations. Analyzes ELO performance, computational efficiency, and training trade-offs. Based on Dennis Soemers’ ML-Agents fork.

Tool for exporting JSON data for further processing with PDFLib POCA.