Scholarlink

<p>In vehicular networks, emerging applications such as autonomous driving and real-time traffic coordination demand ultra-low latency and high-throughput computational capabilities. To meet these requirements, this work presents a distributed task offloading scheme tailored for real-time workflow scheduling in vehicular edge-cloud systems. We propose SACDTO, a Soft Actor-Critic-based algorithm that jointly minimizes task execution delay and computation cost under complex task dependencies. By modeling the workflow as a directed acyclic graph and formulating the problem as a Markov decision process, SACDTO enables vehicles to make offloading decisions in real time based on local observations. The proposed algorithm jointly optimizes task offloading to minimize latency and computation cost while satisfying transmission rate constraints. Moreover, it enables efficient processing on local vehicles and supports decentralized execution. Extensive experiments demonstrate that SACDTO consistently outperforms heuristic baselines in delay reduction and cost efficiency, highlighting its applicability to large-scale, latency-sensitive vehicular systems.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A DRL-based workflow scheduling for cost and delay minimization in vehicular networks

  • 期刊论文

  • 发表日期：

    2025年11月27日

查看全文
  • Luxiu Yin,
  • Wenyu Wu,
  • Jing Huang,
  • Haibo Luo,
  • Wei Liang,
  • Kuanching Li

摘要

In vehicular networks, emerging applications such as autonomous driving and real-time traffic coordination demand ultra-low latency and high-throughput computational capabilities. To meet these requirements, this work presents a distributed task offloading scheme tailored for real-time workflow scheduling in vehicular edge-cloud systems. We propose SACDTO, a Soft Actor-Critic-based algorithm that jointly minimizes task execution delay and computation cost under complex task dependencies. By modeling the workflow as a directed acyclic graph and formulating the problem as a Markov decision process, SACDTO enables vehicles to make offloading decisions in real time based on local observations. The proposed algorithm jointly optimizes task offloading to minimize latency and computation cost while satisfying transmission rate constraints. Moreover, it enables efficient processing on local vehicles and supports decentralized execution. Extensive experiments demonstrate that SACDTO consistently outperforms heuristic baselines in delay reduction and cost efficiency, highlighting its applicability to large-scale, latency-sensitive vehicular systems.