behavior tree

LLM+RL system achieves competitive multi-agent coordination in games

A new research paper introduces a hierarchical control system for multi-agent games, combining large language models (LLMs) for strategic planning with reinforcement learning (RL) for execution. This hybrid approach dem…