Abstract
Unmanned aerial vehicles (UAVs) are increasingly adopted in various applications due to their high mobility and advanced sensing capabilities. However, they also face significant security threats and reliability concerns arising from external adversarial attacks and internal system failures. AI and machine learning techniques have shown promise in detecting security threats and anomalies in UAVs, but their effectiveness heavily depends on high-quality UAV security datasets for training. This thesis presents UAViSec, an open-source simulation platform designed to generate realistic, customizable UAV security scenarios data for cybersecurity and reliability analysis. UAViSec extends the PX4 autopilot and Gazebo simulator with modular attack components that support onboard sensor spoofing, flight-control manipulation, communication interception, and vision-based adversarial attack. The platform enables fine-grained control over attack parameters, customizable patterns, and multi-component fusion attack, allowing researchers to model attack behaviors that closely resemble real-world UAV threats. Using common UAV models and open-source simulated environments, UAViSec demonstrates 26 types of UAV attack effects, including constant, offset, and off attacks on onboard sensors, motor-level control failures, MAVLink command manipulation, and visual signal injection that disrupts object detection. Experimental results confirm that UAViSec produces abnormal flight behavior, system alerts, and detailed log data aligned with expected attack outcomes, offering a scalable and safe alternative to UAV security research.