LLM-Flock: Decentralized Multi-Robot Flocking via Large Language Models and Influence-Based Consensus (2505.06513v1)

Abstract: 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.

Decentralized Multi-Robot Flocking via LLMs

The paper "LLM-Flock: Decentralized Multi-Robot Flocking via LLMs and Influence-Based Consensus" advances the understanding of decentralized control in multi-robot systems by integrating LLMs in the formation strategy. The core of this research lies in addressing the challenges posed by previous attempts to apply LLMs directly for decentralized decision-making, which often resulted in erratic behaviors such as convergence to a centroid or chaotic dispersion. This is primarily due to LLMs' tendency for hallucinated reasoning and lack of cohesive plan alignment across multiple agents.

Methodological Framework

The authors propose a novel framework, LLM-Flock, combining influence-based consensus with LLM-driven local planning. Each robot in the system independently generates a local plan using its onboard LLM, then iteratively refines its planned trajectory considering the influence it holds over neighboring robots. Influence, in this context, is quantified by a robot's communication centrality—those with more neighbors are considered more influential. This enables a decentralized negotiation process that progressively aligns the entire system toward coherent formations, eliminating reliance on centralized control.

Experimental Validation

The framework is validated through extensive simulations and physical deployments involving Crazyflie drones. Simulations incorporate state-of-the-art and open-source LLMs, such as o3-mini, Claude 3.5, Llama3.1-405b, Qwen-Max, and DeepSeek-R1. Results demonstrate improved stability, convergence, and adaptability over standalone LLM-based methods. Notably, the influence-based consensus significantly enhances formation precision and reduces the Procrustes error—a metric reflecting deviations from the desired geometric structure.

Implications and Future Directions

From a theoretical standpoint, LLM-Flock's integration of decentralized consensus with LLM reasoning marks a substantial evolution in multi-agent systems. Practically, the framework enhances real-world viability in applications such as environmental monitoring and search-and-rescue operations by allowing dynamic and adaptive formation control under evolving conditions and objectives.

Future research could build on these findings by exploring hierarchical planning strategies, improving LLM reasoning to mitigate output variability, and enabling on-device inference to extend applicability in communication-limited environments. Overcoming these challenges would potentially enhance LLM-Flock's scalability and efficiency, paving the way for broader adoption in complex multi-robot operational contexts.

In sum, the paper offers significant insights into leveraging LLMs for decentralized multi-robot coordination, providing a robust approach to tackle the often unpredictable nature of decentralized system behavior while demonstrating practical feasibility in both simulated and physical environments. The influence-based consensus protocol emerges as a key innovation, facilitating coherent multi-robot cooperation and advancing the role of LLMs in intelligent systems.