# Building Cooperative Embodied Agents Modularly with Large Language Models

**Authors:** Hongxin Zhang, Weihua Du, Jiaming Shan, Qinhong Zhou, Yilun Du, Joshua B. Tenenbaum, Tianmin Shu, Chuang Gan  
**arXiv:** [2307.02485](https://arxiv.org/abs/2307.02485)  
**Date:** 2023-07-05  
**Summary page:** https://www.summarizepaper.com/en/paper/2307.02485v1/

## Key points

- Authors explore the capabilities of Large Language Models (LLMs) in multi-agent cooperation tasks
- Present a novel framework that utilizes LLMs for multi-agent cooperation in different embodied environments
- Framework enables embodied agents to plan, communicate, and cooperate efficiently for long-horizon tasks
- Recent LLMs like GPT-4 surpass strong planning-based methods and exhibit effective communication without fine-tuning or few-shot prompting
- LLM-based agents that communicate in natural language earn more trust and cooperate effectively with humans
- Highlights potential of LLMs for embodied AI and future research in multi-agent cooperation
- Emphasizes importance of natural language communication for collaboration between human users and AI agents

## Summary

In this paper titled "Building Cooperative Embodied Agents Modularly with Large Language Models," authors Hongxin Zhang, Weihua Du, Jiaming Shan, Qinhong Zhou, Yilun Du, Joshua B. Tenenbaum, Tianmin Shu, and Chuang Gan explore the capabilities of Large Language Models (LLMs) in multi-agent cooperation tasks. While LLMs have shown impressive planning abilities in single-agent embodied tasks across various domains, their capacity for planning and communication in multi-agent settings remains unclear. To address this gap, the authors present a novel framework that utilizes LLMs for multi-agent cooperation and test it in different embodied environments. Their framework enables embodied agents to plan, communicate, and cooperate with other agents or humans efficiently to accomplish long-horizon tasks. The researchers demonstrate that recent LLMs like GPT-4 can surpass strong planning-based methods and exhibit emergent effective communication using their framework without requiring fine-tuning or few-shot prompting. Additionally, the study reveals that LLM-based agents that communicate in natural language can earn more trust and cooperate more effectively with humans. This finding highlights the potential of LLMs for embodied AI and lays the foundation for future research in multi-agent cooperation. The paper provides valuable insights into how LLMs can be leveraged to enhance planning and communication abilities in cooperative multi-agent scenarios. It also emphasizes the importance of natural language communication for effective collaboration between human users and AI agents. Videos related to the project can be found on the project website at https://vis-www.cs.umass.edu/Co-LLM-Agents/.