Timeliness-Oriented Asynchronous Task Offloading in UAV-Edge-Computing Systems
Xiaoqi Qin, Yanlin Li, Nan Ma, Yifan Zhang, Kaifeng Han, Lei Meng, Ping Zhang
Abstract
Unmanned Aerial Vehicles (UAVs) are deployed as digital eyes in the sky to collect and deliver streams of valuable data about a monitored area. To enable autonomous decision making at UAVs, multi-access edge computing (MEC) technology is employed to combat the limited battery and computational capability at UAVs. By offloading computation-intensive tasks to edge server, UAVs can obtain real-time status updates of the observed area by continuous data sampling and processing. The timeliness of obtained status updates is vital to the perception ability at UAVs, which can be characterized by the concept of age of information. In this article, we investigate an age-aware computation offloading scheme in asynchronous MEC systems, where the task generation time instants are heterogeneous among UAVs. The formulated age minimization problem involves binary task offloading variables and continuous computation resource allocation variables, and falls in the category of a stochastic mixed-integer program. Considering the randomness in sequential task generation at each UAV and the time-varying workload at MEC server, we propose a deep reinforcement learning-based solution framework to directly work on the hybrid discrete-continuous action space in the formulated problem. Simulation results show that our proposed scheme outperforms existing strategies in obtaining status updates in a timely manner.