TASK OFFLOADING IN VEHICULAR EDGE COMPUTING USING DEEP REINFORCEMENT LEARNING

Abstract

Vehicular Edge Computing (VEC), a subset of Mobile Edge Computing (MEC), is a promising technology for real-time processing in vehicular networks at the edge of the network. By shifting computationally demanding activities to one-hop-away edge servers, it facilitates the emergence of applications like intelligent transportation systems and autonomous vehicles [3]. There is a combination of problems within the current VEC research area, including a dynamic environment of moving subjects, application task dependency, limited resources on the edge, and constraints on energy efficiency and end-to-end latency. Conventional offloading techniques, like heuristic methods, can not adequately make decisions in such environments in a reasonable time. The goal of this project is to address these challenges by proposing and comparing a standard Feedforward Deep Q-Network (FF-DQN) and an innovative Transformer-Encoder Double Deep Q-Network (TE-DDQN). The proposed methodology incorporates real-world data measurements from the Nvidia Jetson Nano. This work contributes to the advancement of energy and latency-efficient task offloading strategies for the increasingly important domain of vehicular edge computing.