Link: http://arxiv.org/abs/2506.02649v1
PDF Link: http://arxiv.org/pdf/2506.02649v1
Summary: A public safety Unmanned Aerial Vehicle (UAV) enhances situational awarenessin emergency response.
Its agility and ability to optimize mobility andestablish Line-of-Sight (LoS) communication make it increasingly vital formanaging emergencies such as disaster response, search and rescue, and wildfiremonitoring.
While Deep Reinforcement Learning (DRL) has been applied tooptimize UAV navigation and control, its high training complexity, low sampleefficiency, and simulation-to-reality gap limit its practicality in publicsafety.
Recent advances in Large Language Models (LLMs) offer a compellingalternative.
With strong reasoning and generalization capabilities, LLMs canadapt to new tasks through In-Context Learning (ICL), which enables taskadaptation via natural language prompts and example-based guidance, withoutretraining.
Deploying LLMs at the network edge, rather than in the cloud,further reduces latency and preserves data privacy, thereby making themsuitable for real-time, mission-critical public safety UAVs.
This paperproposes the integration of LLM-enabled ICL with public safety UAV to addressthe key functions, such as path planning and velocity control, in the contextof emergency response.
We present a case study on data collection schedulingwhere the LLM-enabled ICL framework can significantly reduce packet losscompared to conventional approaches, while also mitigating potentialjailbreaking vulnerabilities.
Finally, we discuss LLM optimizers and specifyfuture research directions.
The ICL framework enables adaptive, context-awaredecision-making for public safety UAV, thus offering a lightweight andefficient solution for enhancing UAV autonomy and responsiveness inemergencies.
Published on arXiv on: 2025-06-03T09:01:33Z