AI-driven secure offloading for the low-altitude economy with ze-RIS-enabled UAV-MEC

In the evolving landscape of low-altitude economy applications, ensuring secure and efficient task offloading in UAV-assisted mobile edge computing (MEC) has become increasingly critical. This is especially true under the stringent latency, reliability, and security requirements imposed by the 5G and emerging 6G networks. To address these challenges, this paper proposes a deep reinforcement learning (DRL) based zero-energy reconfigurable intelligent surface (ze-RIS) assisted secure task offloading scheme (DzRSS), tailored for 5G-NR-enabled UAV-MEC networks. The proposed scheme combines hybrid offloading modes, adaptive RIS phase optimization, and artificial noise (AN) injection under imperfect channel state information (CSI). A task efficiency and secrecy metric (TESM) is introduced to jointly optimize delay, energy consumption, and secrecy rate, and the problem is formulated as a Markov Decision Process (MDP). An A2C-based DRL agent is used to learn optimal policies that maximize the TESM factor. The system leverages multi-antenna UAV and BS nodes, realistic propagation models with Rician fading and log-normal shadowing, and dynamic beamforming to reinforce legitimate links while suppressing adversaries via null steering. Extensive simulations confirm that DzRSS improves the average TESM by more than 29.13%, increases the secrecy rate by more than 16.01%, and improves the task success rate by 0.86% to 1.87% compared to existing baselines. DzRSS significantly enhances the energy-delay balance, proving its efficiency in safeguarding and optimizing task offloading in challenging adversarial UAV-MEC scenarios.