A Closed-Loop Multi-Agent Framework for Robust Multi-Robot Manipulation

2026-07-08

AI Summary

This work addresses the limited robustness and absence of closed-loop feedback in existing large language model (LLM)-driven multi-robot systems, which hinder reliable high-level task execution—particularly in cross-workspace, contact-intensive scenarios. The authors propose a hierarchical closed-loop multi-agent framework that decomposes high-level instructions into executable subtasks through the coordinated efforts of three specialized agents: planner, operator, and verifier. This architecture adaptively invokes tool-augmented actions and incorporates semantic-level feedback grounded in physical outcomes to iteratively refine execution. By introducing, for the first time, a dedicated multi-agent structure with an integrated closed-loop verification mechanism, the approach effectively bridges the gap between LLM-based reasoning and low-level multi-robot control. Experiments demonstrate significantly improved task success rates in real-world environments, along with strong robustness and generalization across diverse settings, including cross-workspace tasks.

Abstract

Multi-robot systems provide the parallelism and redundancy necessary for long-horizon tasks, while Large Language Models (LLMs) offer the reasoning capabilities to decompose these objectives into actionable plans. However, effectively grounding this high-level reasoning in physical multi-robot execution remains an open challenge. Existing LLM-based approaches fall mainly into two categories: Single-robot methods achieve robust contact-rich manipulation but lack the coordination mechanisms required for tasks spanning multiple workspaces. Current multi-robot frameworks focus on high-level planning, often treating manipulation as an idealized primitive that fails to account for real-world execution uncertainties. To address this, we propose a hierarchical closed-loop agentic LLM-based framework to ensure robust multi-robot manipulation. Our system consists of three specialized agents: the Planning Agent decomposes instructions into allocated sub-tasks, the Manipulation Agent for each robot executes actions via adaptive tool use, and the Verification Agent closes the loop by monitoring physical outcomes and feeding back semantic corrections. Extensive real-world experiments demonstrate that our framework achieves superior success rates, ensures robust adaptability ranging from single to cross workspace manipulation, and offers a generalizable approach for diverse manipulation tasks.

Problem

• multi-robot manipulation
• large language models
• execution grounding
• real-world uncertainty
• cross-workspace coordination

Innovation

• closed-loop multi-agent
• LLM-based manipulation
• robust multi-robot coordination
• adaptive tool use
• semantic feedback