FlockGPT: Guiding UAV Flocking with Linguistic Orchestration

TL;DR

This work tackles scalable drone flocking controlled by natural language. It couples an LLM-driven interface to generate target surfaces via a Python-based Signed Distance Function (SDF) with a distributed flocking controller that assigns drones to surface points and maintains cohesive motion. Validation in Unity with up to 64 simulated drones and real-world Crazyflie experiments shows novice users can craft and recognize six shapes, achieving an average recognition of 80% and up to 93% for certain configurations, while NASA-TLX and UEQ metrics indicate favorable usability and experience. The approach offers an intuitive pathway for large-scale drone shows and teleoperable flocking, with potential extensions to dynamic shapes and richer human–flock interaction modalities.

Abstract

This article presents the world's first rapid drone flocking control using natural language through generative AI. The described approach enables the intuitive orchestration of a flock of any size to achieve the desired geometry. The key feature of the method is the development of a new interface based on Large Language Models to communicate with the user and to generate the target geometry descriptions. Users can interactively modify or provide comments during the construction of the flock geometry model. By combining flocking technology and defining the target surface using a signed distance function, smooth and adaptive movement of the drone swarm between target states is achieved. Our user study on FlockGPT confirmed a high level of intuitive control over drone flocking by users. Subjects who had never previously controlled a swarm of drones were able to construct complex figures in just a few iterations and were able to accurately distinguish the formed swarm drone figures. The results revealed a high recognition rate for six different geometric patterns generated through the LLM-based interface and performed by a simulated drone flock (mean of 80% with a maximum of 93\% for cube and tetrahedron patterns). Users commented on low temporal demand (19.2 score in NASA-TLX), high performance (26 score in NASA-TLX), attractiveness (1.94 UEQ score), and hedonic quality (1.81 UEQ score) of the developed system. The FlockGPT demo code repository can be found at: coming soon

Figures (7)

• Figure 1: Demonstration of FlockGPT setup with a swarm of 8 Crazyflie 2.1 drones.
• Figure 2: Point distribution visualization.
• Figure 3: The diagram illustrates the process of drone allocation on the virtual surface. Green dashed circles show the minimum repulsion radius $r_{safe}$, which is utilized to avoid collisions between the drones. Red arrows represent the velocity vectors $v_i$, directing the UAVs towards the surface defined by SDF. Gray circles on this surface are the points randomly generated before optimization. Green circles are the points optimized for our given number of UAVs. Red crosses indicate the positions recalculated for the drones taking into account their flocking behavior.
• Figure 4: Simulated swarm of drones.
• Figure 5: Examples of executing various commands by a swarm of 64 UAVs in the Unity simulation.