Collaborating Action by Action: A Multi-agent LLM Framework for Embodied Reasoning

2025-04-24

AI Summary

This paper addresses the critical inefficiency of large language model (LLM)-based agents in embodied multi-agent collaboration—particularly a 15% performance degradation under explicit task planning requirements. To tackle this, we propose MINDcraft, a framework featuring a dynamic, natural-language-based interaction protocol, tightly integrated with the Minecraft embodied environment, a modular multi-agent architecture, and dedicated collaboration metrics—moving beyond conventional in-context and imitation learning paradigms. Our empirical analysis, the first of its kind, identifies inefficient inter-agent communication as the primary bottleneck in current LLM-driven embodied collaboration and demonstrates that over-specification in task plans actively impairs coordination. Complementing MINDcraft, we introduce MineCollab—the first standardized benchmark for embodied multi-agent collaboration—designed to evaluate adaptive, open-world reasoning and cooperative problem-solving. Together, MINDcraft and MineCollab advance rigorous, reproducible research on scalable, context-aware collaborative intelligence.

Technology Category

• Multiagent Systems: Coordination and Collaboration
• Intelligent Robots: Embodied AI
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Economics, Online Markets and Human Computation: Architectures and workflows that use LLMs for crowd work
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Collaboration is ubiquitous and essential in day-to-day life -- from exchanging ideas, to delegating tasks, to generating plans together. This work studies how LLMs can adaptively collaborate to perform complex embodied reasoning tasks. To this end we introduce MINDcraft, an easily extensible platform built to enable LLM agents to control characters in the open-world game of Minecraft; and MineCollab, a benchmark to test the different dimensions of embodied and collaborative reasoning. An experimental study finds that the primary bottleneck in collaborating effectively for current state-of-the-art agents is efficient natural language communication, with agent performance dropping as much as 15% when they are required to communicate detailed task completion plans. We conclude that existing LLM agents are ill-optimized for multi-agent collaboration, especially in embodied scenarios, and highlight the need to employ methods beyond in-context and imitation learning. Our website can be found here: https://mindcraft-minecollab.github.io/

Problem

• How LLMs collaborate for embodied reasoning tasks
• Evaluating communication bottlenecks in multi-agent systems
• Optimizing LLM agents for embodied multi-agent collaboration

Innovation

• Multi-agent LLM framework for embodied reasoning
• MINDcraft platform for LLM agents in Minecraft
• Benchmark MineCollab for collaborative reasoning