Edge Service Migration for Vehicular Networks Based on Multi-agent Deep Reinforcement Learning

Abstract

To meet the increasing resource demand of intelligent driving, roadside infrastructure is used to provide communication and computing capabilities to vehicles. Existing studies have leveraged deep reinforcement learning to perform small-scale resource scheduling for vehicles. It is critical to implement large-scale resource scheduling to deal with the high mobility of vehicles. However, this large-scale optimization is confronted with huge state and action space. To overcome this challenge, we propose an edge resource allocation method based on multi-agent deep reinforcement learning to reduce system cost while guarantee the quality of intelligent driving. The proposed method considers both immediate and long-term resource status, which helps to select appropriate base stations and edge servers. Trace driven simulations are performed to validate the efficiency of the proposed method.