关键词: 边缘计算, 自动驾驶, 强化学习, 计算卸载

Abstract: To provide a reasonable real-time scheduling scheme for vehicle tasks with dependencies so that Mobile Edge Computing (MEC) can schedule tasks to be processed in edge nodes for reducing execution time effectively,a real-time scheduling strategy for reasoning tasks in vehicle edge computing was designed.The reasoning process of applications was abstracted as a model based on directed acyclic graphs;the execution order of tasks was defined according to the priority evaluation method,and the appropriate nodes for the corresponding tasks were chosen by Deep Q-learning (DQN).The reasoning tasks were performed according to the results of resource allocation.Experimental results showed that the proposed strategy could effectively reduce the processing delay of reasoning tasks.Compared with the classic algorithms,it had better performance in convergence and response rate.

Key words: edge computing, autonomous driving, reinforcement learning, computation offloading