Edge Collaborative Online Task Offloading Method Based on Reinforcement Learning

Abstract

With the vigorous development of industries such as self-driving, edge intelligence, and the industrial Internet of Things (IoT), the amount and type of data generated are unprecedentedly large, and users’ demand for high-quality services continues to increase. Edge computing has emerged as a new paradigm, providing storage, computing, and networking resources between traditional cloud data centers and end devices with solid timeliness. Therefore, the resource allocation problem in the online task offloading process is the main area of research. It is aimed at the task offloading problem of delay-sensitive customers under capacity constraints in the online task scenario. In this paper, a new edge collaborative online task offloading management algorithm based on the deep reinforcement learning method OTO-DRL is designed. Based on that, a large number of simulations are carried out on synthetic and real data sets, taking obstacle recognition and detection in unmanned driving as a specific task and experiment. Compared with other advanced methods, OTO-DRL can well realize the increase in the number of tasks requested by mobile terminal users in the field of edge collaboration while guaranteeing the service quality of task requests with higher priority.