FRSICL: LLM-Enabled In-Context Learning Flight Resource Allocation for Fresh Data Collection in UAV-Assisted Wildfire Monitoring

TL;DR

The paper addresses minimizing data staleness in UAV-assisted wildfire monitoring by jointly optimizing data collection schedules and UAV velocity. It introduces FRSICL, an online flight resource allocation framework that leverages LLM-enabled In-Context Learning to generate decisions from natural-language task descriptions and environment feedback, deployed via an edge-hosted LLM. Simulation results show FRSICL achieves significantly lower average AoI than PPO, BCD, and nearest-neighbor baselines, while maintaining stable velocity profiles. The approach offers training-free adaptability, interpretability, and a clear path toward multi-UAV extensions, with practical considerations for edge efficiency and security.

Abstract

Uncrewed Aerial Vehicles (UAVs) play a vital role in public safety, especially in monitoring wildfires, where early detection reduces environmental impact. In UAV-Assisted Wildfire Monitoring (UAWM) systems, jointly optimizing the data collection schedule and UAV velocity is essential to minimize the average Age of Information (AoI) for sensory data. Deep Reinforcement Learning (DRL) has been used for this optimization, but its limitations-including low sampling efficiency, discrepancies between simulation and real-world conditions, and complex training make it unsuitable for time-critical applications such as wildfire monitoring. Recent advances in Large Language Models (LLMs) provide a promising alternative. With strong reasoning and generalization capabilities, LLMs can adapt to new tasks through In-Context Learning (ICL), which enables task adaptation using natural language prompts and example-based guidance without retraining. This paper proposes a novel online Flight Resource Allocation scheme based on LLM-Enabled In-Context Learning (FRSICL) to jointly optimize the data collection schedule and UAV velocity along the trajectory in real time, thereby asymptotically minimizing the average AoI across all ground sensors. Unlike DRL, FRSICL generates data collection schedules and velocities using natural language task descriptions and feedback from the environment, enabling dynamic decision-making without extensive retraining. Simulation results confirm the effectiveness of FRSICL compared to state-of-the-art baselines, namely Proximal Policy Optimization, Block Coordinate Descent, and Nearest Neighbor.

Figures (6)

• Figure 1: A representative UAV-assisted wildfire monitoring scenario, illustrating sensor data collection via LoS wireless links between the UAV and ground sensors.
• Figure 2: Overview of the proposed FRSICL framework: The UAWM system interacts with an edge-hosted LLM using structured prompts. The LLM receives logged environmental data (e.g., AoI, channel conditions, and UAV location) and generates data collection schedules and velocities. A feedback loop records system performance and enables continuous adjustment to minimize average AoI.
• Figure 3: Data communication protocol for the proposed FRSICL, including prompting the LLM, UAV movement and sensor selection, sensor transmission, acknowledgement and update, and iteration.
• Figure 4: Performance analysis of FRSICL in different scenarios with 10 ground sensors (a) Performance of FRSICL against BCD, and PPO baselines over time steps
• Figure 5: Comparative velocity trajectories during UAV-assisted wildfire monitoring. FRSICL achieves lower average velocity than PPO and smoother velocity than Nearest Neighbor.