​Abstract: This article presents a security-aware Edge Artificial Intelligence framework designed to enhance autonomous decision-making in Unmanned Aerial Systems. As aerial vehicles increasingly rely on onboard artificial intelligence to interpret sensor data, plan flight trajectories, and respond to environmental conditions, they are exposed to cyber-physical threats such as data manipulation, Global Navigation Satellite System (GNSS) spoofing, and adversarial visual perturbations. The proposed framework integrates lightweight deep learning models optimized for embedded processors with a multilayer cybersecurity architecture that includes real-time integrity verification of sensor streams, adversarial robustness modules, and secure decision-validation routines. Experimental evaluation on a quadrotor platform demonstrates that the system preserves autonomy performance while significantly reducing the risk of unsafe commands under adversarial conditions, achieving over an 85 % reduction in unsafe actions with minimal false alarms and negligible impact on mission continuity. These findings highlight the importance of combining Edge Artificial Intelligence with embedded security mechanisms to ensure the resilience, safety, and reliability of autonomous unmanned systems operating in contested or communication-limited environments.

Keywords: Edge artificial intelligence, Unmanned aerial systems, UAV cybersecurity, Autonomous decision-making, Embedded AI, Adversarial robustness, Sensor integrity monitoring.

__________________________________________________________________________________________