📄 中文摘要
随着现实场景对多个具身代理在动态环境中协作的需求日益增加，许多任务超出了单个代理的能力。大型语言模型（LLMs）的最新进展使得通过推理、规划和自然语言交流实现高层次的认知协调成为可能。然而，现有基准难以对这种合作如何出现、展开以及如何促进具身多代理系统的任务成功进行细致分析。EmCoop作为一个基准框架被提出，用于研究基于LLM的具身多代理系统中的合作。该框架将高层认知层与低层具身交互层分离，从而提供了更清晰的分析视角。
EmCoop：一个用于大语言模型代理之间体现合作的框架和基准
📄 English Summary
EmCoop: A Framework and Benchmark for Embodied Cooperation Among LLM Agents
The increasing demand for multiple embodied agents to collaborate in dynamic environments arises from the limitations of single-agent capabilities in real-world scenarios. Recent advancements in large language models (LLMs) facilitate high-level cognitive coordination through reasoning, planning, and natural language communication. However, existing benchmarks struggle to provide fine-grained analyses of how such collaboration emerges, unfolds, and contributes to task success in embodied multi-agent systems. EmCoop is introduced as a benchmark framework aimed at studying cooperation in LLM-based embodied multi-agent systems. This framework separates the high-level cognitive layer from the low-level embodied interaction layer, offering a clearer perspective for analysis.