Learning-Based Detection of Incorrect Control Parameter Configurations for Robotic Aerial Vehicle

Robotic Aerial Vehicle (RAV) provides individual modifications of control parameters to perform various missions in different environments. The system's high operability and adaptability depend on the correct configuration of the control parameters. To prevent users from misconfiguring parameters and causing control problems, RAV developers such as PX4 and Ardupilot usually set legal boundaries for each control parameter. The flight controller will trigger an alarm for control parameters that exceed these boundaries. However, recent studies have discovered that certain parameter combinations, even within these legal limits, can still result in serious consequences for RAV s, including performance degradation, system failures, and even crashes. Current practices, which often rely solely on parameter range settings, are insufficient in detecting such incorrect configurations. Therefore, there is an emerging need for a viable method to detect incorrect parameter configurations in RAVs. This paper presents a machine learning classifier-based approach to detect control parameters that can induce operational issues under specific configurations. Experimental results comparing five different machine learning methods�K�Nearest Neighbors (KNN), Support Vector Machine (SVM), Decision Trees (DT), Random Forests (RF), and Naive Bayes (NB)�in both simulated environments and real-world scenarios demonstrate the effectiveness of the proposed approach, achieving 94% accuracy, 93% precision, 99% recall, and 96% F1-scores.