The arXiv preprint LLM-Flock (2505.06513) explores a new role for large language models: as decentralized decision-makers for multi-robot formation control. The authors note that LLMs have strong reasoning and planning capabilities, which has prompted interest in using them to coordinate robot swarms without a central controller.
To address this, the paper proposes an approach that combines influence-based consensus with collision avoidance. The title and abstract suggest that the method is designed to let each robot make local decisions using LLM-driven reasoning while still maintaining coherent group behavior and avoiding collisions.
It is worth noting that the available abstract is truncated, so the specific challenges of directly applying LLMs to this task are not fully detailed in the source. As a preprint, the work has not yet undergone peer review, and the version cited is a replacement, indicating ongoing revisions.