Hierarchical Control in Multi-Agent Games: LLM-based Planning and RL Execution

The article presents a hierarchical architecture that integrates a pretrained large language model (LLM) as a centralized strategic controller with specialized reinforcement learning (RL) skill policies for multi-agent environments. In a competitive 2v2 King of the Hill setting, this LLM+RL system achieved performance comparable to behavior tree baselines, with a win rate of 46.4% versus 51.5%, while outperforming traditional Flat RL approaches. This approach highlights the potential for LLMs to enhance multi-agent coordination and perceived adaptability, reducing the need for manual rule engineering in complex decision-making scenarios.