NAVER's DroneCATS Finds MLLMs Fly, Fail to Declare Arrival

NAVER Cloud's DroneCATS benchmark evaluates multimodal large language models (MLLMs) for zero-shot drone control, showing smaller open models navigate better but fail protocol discipline, while frontier models like GPT-5 perform inconsistently, especially in multi-drone commanding.

This Wire brief sits within Fusion42's coverage of Robotics, Drones & UAV.