TD3-Based Joint Trajectory-Power-Beamforming Optimization for Secure and Covert UAV-Aided Vehicular Edge Computing

Abstract

Unmanned aerial vehicle (UAV)-aided vehicular edge computing (VEC) faces dual challenges of security against eavesdropping and covertness against detection, especially in high-mobility vehicle-to-everything scenarios. This paper presents a UAV-aided multi-vehicle secure-covert edge-offloading system that incorporates comprehensive channel, covert-communication, and computational models. To achieve secure, covert, and low-latency adaptive edge computing in the presence of eavesdroppers (Eves), we propose a twin-delayed deep deterministic policy gradient-based deep reinforcement learning scheme that jointly optimizes UAV trajectory, transmission power, beamforming, and task offloading, while ensuring covertness via a novel Rician-aware power constraint derived from a second-order Padé approximation. Simulations show our scheme achieves a 15% higher cumulative reward and reduces system latency by 17% compared to the second-best baseline, while maintaining covertness under dynamic Eve locations. Meanwhile, the scheme also demonstrates robustness to channel estimation errors, maintaining over 90% performance even under moderate error conditions.