LLM-Flock: Decentralized Multi-Robot Flocking via Large Language Models and Influence-Based Consensus

TL;DR

This work tackles decentralized multi-robot flocking by addressing the instability of uncoordinated LLM planning. It introduces LLM-Flock, a framework that couples per-robot LLM-based local planning with an influence-based consensus to coherently align plans without central control. The approach is validated across multiple closed- and open-source LLM backends, in both simulations and real Crazyflie drones, showing improved stability, convergence, and adaptability over prior methods. The results demonstrate the potential of structured, language-model–guided coordination for scalable, decentralized multi-robot formation control, while also outlining practical limitations and directions for future enhancement.

Abstract

Large Language Models (LLMs) have advanced rapidly in recent years, demonstrating strong capabilities in problem comprehension and reasoning. Inspired by these developments, researchers have begun exploring the use of LLMs as decentralized decision-makers for multi-robot formation control. However, prior studies reveal that directly applying LLMs to such tasks often leads to unstable and inconsistent behaviors, where robots may collapse to the centroid of their positions or diverge entirely due to hallucinated reasoning, logical inconsistencies, and limited coordination awareness. To overcome these limitations, we propose a novel framework that integrates LLMs with an influence-based plan consensus protocol. In this framework, each robot independently generates a local plan toward the desired formation using its own LLM. The robots then iteratively refine their plans through a decentralized consensus protocol that accounts for their influence on neighboring robots. This process drives the system toward a coherent and stable flocking formation in a fully decentralized manner. We evaluate our approach through comprehensive simulations involving both state-of-the-art closed-source LLMs (e.g., o3-mini, Claude 3.5) and open-source models (e.g., Llama3.1-405b, Qwen-Max, DeepSeek-R1). The results show notable improvements in stability, convergence, and adaptability over previous LLM-based methods. We further validate our framework on a physical team of Crazyflie drones, demonstrating its practical viability and effectiveness in real-world multi-robot systems.

Figures (14)

• Figure 1: LLM-Flock framework for decentralized multi-robot flocking. Each robot first uses an onboard LLM to generate a candidate plan from a shared prompt. Then robots exchange plans locally and iteratively refine them through an influence-based consensus protocol. Motion execution occurs in parallel, with each robot’s LLM deciding the waypoint location at each step, enabling coordination without centralized control.
• Figure 2: Comparison of formation results with and without influence-based consensus. (a)-(d) show robot trajectories using influence-based plan consensus with three robots forming triangle formations, and (e)-(h) show the trajectories without consensus.
• Figure 3: Qualitative results demonstrating LLM-Flock on diverse formation tasks. (a)-(d) show the progression of robot trajectories as eight robots form a square formation using influence-based plan consensus. (e)-(h) show the progression for ten robots forming a circle formation.
• Figure 4: Procrustes shape error over time across different formation tasks and LLM backends. Lines show the mean Procrustes error across 10 randomized trials, with shaded regions representing the 95% confidence intervals.
• Figure 5: Snapshots of five Crazyflie drones forming a circle using proposed LLM-Flock framework.