Researchers propose LLM-Foraging, a decentralized swarm robot controller that uses large language models to make tactical decisions without requiring retraining when deployment conditions change.

LLM-Foraging augments the central-place foraging algorithm (CPFA) state machine with an LLM tactical decision-maker at three structured decision points: post-deposit, central-zone arrival, and search starvation. Each robot runs its own LLM client and queries it using only locally observable state.

Unlike traditional approaches using genetic algorithms or reinforcement learning, LLM-Foraging is training-free at deployment and transfers across configurations without re-optimization, because the LLM serves as a general decision policy rather than parameters fitted to a single configuration.

In Gazebo evaluation with TurtleBot3 robots across 36 configurations spanning team sizes of 4 to 10 robots, arena sizes from 6x6 to 10x10 meters, and three resource distributions (clustered, powerlaw, random), LLM-Foraging collects more resources than the GA-tuned CPFA baseline and is more consistent across configurations.