Date of Award

8-23-2024

Publication Type

Thesis

Degree Name

M.A.Sc.

Department

Electrical and Computer Engineering

Keywords

CAVs;CVIS;DDPG;DRL;Reverse Offloading;VEC

Supervisor

Ning Zhang

Creative Commons License

Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.

Abstract

Research on autonomous driving has attracted attention from scientists all over the world. The communication among autonomous vehicles and other elements on the road, such as other vehicles, infrastructure, and pedestrians, constitutes crucial supplements to autonomous driving, namely connected autonomous vehicles (CAVs). In the thesis, we discuss the development of autonomous vehicles and CAVs and conduct a comprehensive literature review on communication architecture regarding CAVs. Although the advent of cooperative vehicle-infrastructure system (CVIS) and vehicular edge computing (VEC) brings the futuristic transportation system into reality, they also bring the challenge of huge data processing demand. Fortunately, with reverse offloading, the soaring computing capabilities of CAVs can be utilized to address the scarcity of computing power. In our proposed method, we adopt a reverse offloading framework in a VEC network, aiming to reduce the system latency through intelligent task partition and resource allocation. We first formulate the reverse offloading problem as a Markov decision process (MDP), considering the time-varying status of channel quality, task queueing status, and task load. Then, we propose a Deep Reinforcement Learning (DRL) algorithm to learn optimal decisions by interacting with the environment. Specifically, a deep deterministic policy gradient (DDPG) based algorithm is proposed to make task partition and allocation decision in a constrained continuous manner with the help of Softmax function. Simulation results demonstrate that the proposed approach can significantly reduce the system latency compared to the three baseline schemes. Notably, when facing the longer traffic rush time, the advantages of the proposed method are progressively growing.

Available for download on Saturday, February 22, 2025

Share

COinS