Towards Efficient LLM Grounding for Embodied Multi-Agent Collaboration

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Zhang, Yang, Yang, Shixin, Bai, Chenjia, Wu, Fei, Li, Xiu, Wang, Zhen, Li, Xuelong
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866916975184183296
author Zhang, Yang
Yang, Shixin
Bai, Chenjia
Wu, Fei
Li, Xiu
Wang, Zhen
Li, Xuelong
author_facet Zhang, Yang
Yang, Shixin
Bai, Chenjia
Wu, Fei
Li, Xiu
Wang, Zhen
Li, Xuelong
contents Grounding the reasoning ability of large language models (LLMs) for embodied tasks is challenging due to the complexity of the physical world. Especially, LLM planning for multi-agent collaboration requires communication of agents or credit assignment as the feedback to re-adjust the proposed plans and achieve effective coordination. However, existing methods that overly rely on physical verification or self-reflection suffer from excessive and inefficient querying of LLMs. In this paper, we propose a novel framework for multi-agent collaboration that introduces Reinforced Advantage feedback (ReAd) for efficient self-refinement of plans. Specifically, we perform critic regression to learn a sequential advantage function from LLM-planned data, and then treat the LLM planner as an optimizer to generate actions that maximize the advantage function. It endows the LLM with the foresight to discern whether the action contributes to accomplishing the final task. We provide theoretical analysis by extending advantage-weighted regression in reinforcement learning to multi-agent systems. Experiments on Overcooked-AI and a difficult variant of RoCoBench show that ReAd surpasses baselines in success rate, and also significantly decreases the interaction steps of agents and query rounds of LLMs, demonstrating its high efficiency for grounding LLMs. More results are given at https://embodied-read.github.io
format Preprint
id arxiv_https___arxiv_org_abs_2405_14314
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Towards Efficient LLM Grounding for Embodied Multi-Agent Collaboration
Zhang, Yang
Yang, Shixin
Bai, Chenjia
Wu, Fei
Li, Xiu
Wang, Zhen
Li, Xuelong
Artificial Intelligence
Computation and Language
Machine Learning
Multiagent Systems
Robotics
Grounding the reasoning ability of large language models (LLMs) for embodied tasks is challenging due to the complexity of the physical world. Especially, LLM planning for multi-agent collaboration requires communication of agents or credit assignment as the feedback to re-adjust the proposed plans and achieve effective coordination. However, existing methods that overly rely on physical verification or self-reflection suffer from excessive and inefficient querying of LLMs. In this paper, we propose a novel framework for multi-agent collaboration that introduces Reinforced Advantage feedback (ReAd) for efficient self-refinement of plans. Specifically, we perform critic regression to learn a sequential advantage function from LLM-planned data, and then treat the LLM planner as an optimizer to generate actions that maximize the advantage function. It endows the LLM with the foresight to discern whether the action contributes to accomplishing the final task. We provide theoretical analysis by extending advantage-weighted regression in reinforcement learning to multi-agent systems. Experiments on Overcooked-AI and a difficult variant of RoCoBench show that ReAd surpasses baselines in success rate, and also significantly decreases the interaction steps of agents and query rounds of LLMs, demonstrating its high efficiency for grounding LLMs. More results are given at https://embodied-read.github.io
title Towards Efficient LLM Grounding for Embodied Multi-Agent Collaboration
topic Artificial Intelligence
Computation and Language
Machine Learning
Multiagent Systems
Robotics
url https://arxiv.org/abs/2405.14314