Towards Efficient LLM Grounding for Embodied Multi-Agent Collaboration
This work addresses the problem of inefficient LLM grounding for embodied multi-agent tasks, offering a method that reduces computational costs and improves coordination, though it appears incremental as it builds on existing self-reflection and reinforcement learning techniques.
The paper tackles the challenge of efficiently grounding large language models (LLMs) for embodied multi-agent collaboration by proposing a framework with Reinforced Advantage feedback (ReAd) to reduce excessive LLM queries. Experiments on Overcooked-AI and RoCoBench show that ReAd surpasses baselines in success rate and significantly decreases interaction steps and query rounds.
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