Machine Learning Based Joint Offloading and Trajectory Design in UAV Based MEC System for IoT Devices

In this paper, we consider an unmanned aerial vehicle (UAV)-based mobile edge computing (MEC) system for Internet of Things devices (IoTDs), in which the mobile UAV equipped with computing resources provides service for all IoTDs in the system and each IoTD also has a certain amount of computing power. Each IoTD has a computing task to be completed and the computing task will be offloaded to the UAV or implemented locally. After the UAV completes the service for any IoTD, it proceeds to the next IoTD, until that all IoTDs in the system have been served. The UAV can collect all information related to all IoTDs before taking off, such as the data size of the computing task and the coordinate of the locations of all IoTDs. Our goal is to minimize the total delay of the system (i.e., flying delay, transmission delay, local computing delay and UAV-aided edge computing delay) through joint optimization of the flying trajectory of the UAV and the ratio of the offloading tasks. Aiming at this problem, a machine learning framework based on Q-learning algorithm is proposed so as to minimize the total delay of the system. Simulation results show that the proposed algorithm can achieve up to 15.7% and 38.0% gain compared to the random algorithm with dynamic selection of task offloading mode and the Q-learning algorithm with fixed selection of task offloading mode.