评估多模态大语言模型作为无人机控制的通用视觉-语言-行动智能体:指挥、接近、跟踪与搜索
TLDR
提出DroneCATS-Agent与DroneCATS基准,评估MLLM作为无人机控制智能体,发现小模型能导航但不会声明到达。
评分理由
The paper presents a novel benchmark and architecture for studying MLLMs as drone-control agents, with a useful focus on scaling down to small models and identifying failure modes. However, the abstract is truncated and does not clarify whether evaluations are real-world or simulated, and the work is not centered on world models.
Read-first 评分解释
综合优先阅读分 14.9,由主题、引用、图谱、方法、可复现性和近期性等信号加权得到。 原始总分保留为 0。
研究版图角色
前沿论文
排序敏感性
稳定性:volatile;排名波动范围:27。