.png){kind=icon}
.png){kind=icon}
Return
Dynamic and Adaptive Task Offloading for UAV-Enabled MEC Systems
.png){kind=icon}
PRE
AI
DOI：10.1016/j.future.2026.108505.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Unmanned Aerial Vehicles (UAVs) are increasingly used as mobile edge computing (MEC) platforms in dynamic environments such as disaster zones, where ground infrastructure may be unavailable or damaged. While UAV-MEC systems can provide flexible computation offloading, many existing approaches either overlook the joint energy constraints of UAVs and end devices or depend on baseline deep reinforcement learning (DRL) algorithms such as DDPG and TD3, which often suffer from instability and poor adaptability in continuous action spaces. This work proposes a UAV-enabled MEC system that jointly optimizes UAV trajectory, wireless communication, computation offloading, and the energy consumption of both UAVs and user devices. The problem is formulated as a multi-objective optimization balancing latency and energy efficiency, two tightly coupled factors that directly affect system performance. To address this problem, we develop a SAC-based constrained learning framework (SACDC) tailored for UAV-enabled MEC. SACDC builds upon the standard Soft Actor–Critic framework and does not modify its underlying learning principle; instead, it integrates domain-specific mechanisms such as state normalization, action correction, entropy tuning, and explicit latency–energy monitoring to ensure stable and feasible control under dynamic UAV-MEC conditions. Simulation results show that SACDC improves convergence speed by nearly 30% over DDPG and TD3, reduces average latency to about 2.2 s (compared to 2.7–2.8, s), and lowers energy consumption by up to 80–90%. These gains hold across varying numbers of users, computing resources, and communication bandwidths. Overall, SACDC enables efficient and reliable task offloading in UAV-enabled MEC, supporting mission-critical applications under highly dynamic conditions.
Keywords:
UAV-enabled MEC
task offloading
deep reinforcement learning
energy efficiency
trajectory optimization
Journal
F
IF:
0
Papers:
642
Citations:
0