Multi-Agent Training Framework adds Opponent-Learning Awareness capability

AI Impact Summary

This change introduces opponent-learning awareness into the learning process, allowing agents to anticipate opponents' adaptation during policy updates. It increases training complexity and non-stationarity, necessitating changes to data collection, rollout management, and evaluation to ensure stable convergence and reliable metrics. Without proper retraining and monitoring, teams may see unstable performance or longer training cycles, but with it, robustness against adaptive opponents should improve.