Age of Task Oriented Offloading Design for UAV-assisted Mobile Edge Computing Systems

To alleviate the computational resource scarcity of wireless ground users (GUs), unmanned aerial vehicles (UAVs) are used to assist mobile edge computing (MEC). This is envisaged as an effective way for GUs to offload a fraction of their computationally-intensive tasks to edge MEC servers for fast execution. However, the freshness of task execution is crucial for time-sensitive applications. Out-of-date results may be meaningless and lead to erroneous decisions. Therefore, the paper employs the concept of age-of-tasks (AoT) to quantify the information freshness during task execution. Multiple UAVs’ relay service assignment, trajectory, as well as GUs’ local transmission power and local computing frequency, should be jointly optimized to minimize AoT subject to energy constraints. This is challenging due to the non-convex mixed-integer nonlinear programming (MINLP) nature. The paper proposes a deep deterministic policy gradient (DDPG) algorithm combined with successive convex approximation (SCA) optimization. The service assignment is solved in advance to reduce training complexity using the DDPG approach. Then, other variables such as UAVs’ trajectory, task offloading fraction, power, and computing frequency can be solved iteratively using SCA. Numerical results reveal the effectiveness and superiority of the proposed DDPG-SCA combined algorithm in achieving the minimum AoT compared with the counterpart offloading rate maximization scheme.