RisConFix: LLM-based Automated Repair of Risk-Prone Drone Configurations

RISCONFIX: LLM-BASED AUTOMATED REPAIR OF RISK-PRONE DRONE CONFIGURATIONS Liping Han School of Computer Science Nanjing University of Posts and Telecommunications Nanjing, China liping@njupt.edu.cn Tingting Nie School of Computer Science Nanjing University of Posts and Telecommunications Nanjing, China nietingting0819@gmail.com Le Yu School of Computer Science Nanjing University of Posts and Telecommunications Nanjing, China yulele08@njupt.edu.cn Mingzhe Hu School of Computer Science Nanjing University of Posts and Telecommunications Nanjing, China hmz@njupt.edu.cn Tao Yue School of Computer Science and Engineering Beihang University Beijing, China yuetao@buaa.edu.cn December 9, 2025 ABSTRACT Flight control software is typically designed with numerous configurable parameters governing multi- ple functionalities, enabling flexible adaptation to mission diversity and environmental uncertainty. Although developers and manufacturers usually provide recommendations for these parameters to ensure safe and stable operations, certain combinations of parameters with recommended values may still lead to unstable flight behaviors, thereby degrading the drone’s robustness. To this end, we pro- pose a Large Language Model (LLM) based approach for real-time repair of risk-prone configurations (named RisConFix) that degrade drone robustness. RisConFix continuously monitors the drone’s operational state and automatically triggers a repair mechanism once abnormal flight behaviors are detected. The repair mechanism leverages an LLM to analyze relationships between configuration parameters and flight states, and then generates corrective parameter updates to restore flight stability. To ensure the validity of the updated configuration, RisConFix operates as an iterative process; it continuously monitors the drone’s flight state and, if an anomaly persists after applying an update, automatically triggers the next repair cycle. We evaluated RisConFix through a case study of ArduPi- lot (with 1,421 groups of misconfigurations). Experimental results show that RisConFix achieved a best repair success rate of 97% and an optimal average number of repairs of 1.17, demonstrating its capability to effectively and efficiently repair risk-prone configurations in real time. 1 Introduction Drones have become widely used in diverse tasks, such as environmental monitoring and emergency response. Un- like conventional smart devices directly operated by humans, drones are controlled by flight control software that autonomously manages flight operations. To support diverse missions and operate under uncertain environmental conditions, modern flight control software (e.g., PX4 [1], and ArduPilot [2]) provide a large number of configurable arXiv:2512.07122v1 [cs.SE] 8 Dec 2025 A PREPRINT - DECEMBER 9, 2025 parameters governing multiple functionalities, for example, flight control (attitude, position, and power system regula- tion), mission execution (flight mode), hardware adaptation (sensor calibration, actuator mapping, and communication link configuration) etc. However, recent studies have shown that certain combinations of these parameters may induce undesirable flight behaviors, including unstable oscillations and even crashes. In this paper, we refer to this type of improper configuration as risk-prone configurations, which poses a severe threat to drone robustness. For example, inappropriate tuning of attitude controller gains may cause oscillations or delays in response, while suboptimal throttle and pitch limit settings can result in altitude divergence or sudden thrust loss. To ensure configuration reliability, prior studies have primarily focused on detecting misconfigurations before deploy- ment. For instance, LGDFUZZER [3] and ICSEARCHER [4] employ fuzzing and learning-guided search techniques to detect incorrect configurations. While these approaches have proven effective in detecting potential risks, they remain constrained to offline analysis or pre-deployment testing. In practice, however, it is nearly impossible to exhaustively capture all risk-prone configurations during testing. This limitation highlights the necessity of an online and adaptive mechanism capable of detecting and repairing risk-prone configurations in real time, thereby maintaining robust flight behavior during dynamic operations. To fill this gap, we propose RisConFix, a Large Language Model (LLM) based automated repair framework for risk-prone drone configurations. RisConFix continuously monitors the drone’s operational state and automatically triggers a repair mechanism once abnormal flight behaviors are detected. The repair mechanism leverages an LLM to generate corrective configuration updates that restore flight stability in real time. Specifically, RisConFix consists of two main phases: 1) Runtime Anomaly Monitor, which continuously collects flight data (e.g., attitude, velocity, control outputs) and detects abnormal flight behaviors via a wireless comm

This content is AI-processed based on open access ArXiv data.