2.118

影响因子

    高级检索

    多循环嵌套的大语言模型多智能体指挥控制过程

    Multi-loop Nested LLM-based Multi-agent Command and Control Processes

    • 摘要: 聚焦于闭合指挥控制链路中大语言模型多智能体系统,旨在探索大模型多智能体在指挥控制链路中的关键技术,提出多循环嵌套的大语言模型多智能体指挥控制模型,为突破无人化指挥控制、自主协同打下坚实的理论和技术基础。充分考虑到大语言模型问答式的交互模式,提出P (R)EA&OODA多循环嵌套的大模型多智能体指挥控制过程模型,以大模型系统提示词为切入点,提出大模型指挥控制智能体框架,设计指挥员和行动员两种 “拟人化” 大模型智能体,设计实现基于物理引擎的文本指挥控制仿真任务 “追击游戏” 从多个维度对系统进行评估和分析,验证模型的有效性。

       

      Abstract: Focusing on the multi-agent system of LLMs in closed C2 links, the key technical issues of LLM multi-agents in C2 links is aimed to be explored , and a multi-loop nested LLM multi-agent C2 model is pro-posed, a solid theoretical and technical foundation for breaking through unmanned C2 and autonomous colla-boration is laid. In full consideration of the question-and-answer interaction mode of LLMs, a P(R)EA&OODA multi-loop nested LLM multi-agent C2 process model is proposed. Taking the system prompt as the core entry point, a framework for LLMC2 agents is proposed. Two types of "personified" LLM agents of commanders and operators are designed. A text-based C2 simulation task "pursuit game" based on a physics engine is designed and implemented to verify the effectiveness of the model, and the system is evaluated and analyzed from multiple dimensions.

       

    /

    返回文章
    返回