Computation offloading strategy for vehicle with deep reinforcement learning
Name
Tzu-Liang Huang
Major
Data Science
Class
2023
About
Tzuliang Huang
th270
Signature Work Project Overview
Task offloading decision-making is essential in vehicular edge computing, as it facilitates the fulfillment of intricate vehicle task demands while minimizing network resource conflicts and usage. In this paper, we introduce a deep reinforcement learning-based task offloading decision algorithm called Deep Reinforcement learning based Offloading Decision (DROD), specifically designed for Vehicle Edge Computing (VEC). By combining the Manhattan mobility model and the intelligent driver model, we create a realistic simulation environment that allows vehicles to make offloading decisions based on their local states.
Our method tackles a complex environmental challenge that encompasses vehicle mobility, signal attenuation, and resource utilization, aiming to optimize the task completion rate. DROD utilizes a Markov decision process to represent the interactions between vehicles and mobile edge computing (MEC) servers. To facilitate iterative training and optimal decision-making, a deep learning algorithm, referred to as DRL, is implemented.
DRL operates on a single-vehicle-observable state space and utilizes a replay buffer for improved learning efficiency. We conduct experiments to evaluate DROD’s performance, demonstrating the feasibility of using DRL for effective task offloading in complex VEC environments.
