A Dynamic Model of Multi-Agent Reinforcement Learning Based on the Framework of Game Theory

Modeling multi-agent learning dynamics has long been a significant research topic. A multi-agent Q-Learning model based on game theory is proposed. In large systems, agents interact using symmetric normal-form games and they employ Boltzmann exploration to learn optimal strategies, dynamic equations capture this learning process. Simulations show the model accurately describes behavior evolution and agents can learn to converge to Nash equilibrium strategies, validating model stability.