Scholarlink

<p>Multi-access Edge Computing (MEC) is a key enabler for supporting latency-sensitive and compute-intensive applications in autonomous vehicles by integrating edge computing resources with vehicular Ad Hoc Networks (VANETs). Existing strategies for task execution, such as local-only, static offloading to RSUs, and greedy minimum-latency, exhibit critical flaws, including low scalability in dense traffic, imbalanced RSU utilisation, numerous handovers, and significant energy consumption. These challenges underscore the need for dynamic, intelligent multi-objective decision-making frameworks that enable real-time trade-offs. In this paper, we first identify challenging vehicular mobility and network conditions. Then we design DRLO-VANET, a framework for dynamic local execution vs. MEC offloading decision-making based on deep reinforcement learning. A novel framework for integrating NS-3 and ns3-gym, enabling online state-action-reward learning with sophisticated model-free DRL algorithms such as Deep Q-Networks (DQN) and Soft Actor-Critic (SAC), is presented. DRLO-VANET introduces a global state space that includes vehicular density, RSU load, task size, and channel quality, and jointly optimises latency, energy consumption, task completion ratio, RSU utilisation rate, and handover overhead. Through extensive simulations, experimental results reveal that DRLO-VANET decreases the task execution latency by up to 40%, saves energy (30%–35%), and enhances the task completion ratio to over 90% with medium density, and decreases the handover frequency by nearly 50% analysis compared to the baseline methods. As a result, these results verify that policies activated by DRL outperform other energy management strategies by balancing short-term responsiveness with long-term system stability. We proposed a scalable, adaptive, and efficient framework with real-time task offloading capabilities for vehicular tasks. Its performance was proven through initial experiments and simulations, making it suitable for new autonomous transport systems that require high reliability and responsiveness.</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

DRLO-VANET: a deep reinforcement learning-based offloading framework for low-latency and energy-efficient task execution in VANETs

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月30日

查看全文
  • Sadineni Neelima,
  • S. Rama Sree,
  • N. Ramakrishnaiah

摘要

Multi-access Edge Computing (MEC) is a key enabler for supporting latency-sensitive and compute-intensive applications in autonomous vehicles by integrating edge computing resources with vehicular Ad Hoc Networks (VANETs). Existing strategies for task execution, such as local-only, static offloading to RSUs, and greedy minimum-latency, exhibit critical flaws, including low scalability in dense traffic, imbalanced RSU utilisation, numerous handovers, and significant energy consumption. These challenges underscore the need for dynamic, intelligent multi-objective decision-making frameworks that enable real-time trade-offs. In this paper, we first identify challenging vehicular mobility and network conditions. Then we design DRLO-VANET, a framework for dynamic local execution vs. MEC offloading decision-making based on deep reinforcement learning. A novel framework for integrating NS-3 and ns3-gym, enabling online state-action-reward learning with sophisticated model-free DRL algorithms such as Deep Q-Networks (DQN) and Soft Actor-Critic (SAC), is presented. DRLO-VANET introduces a global state space that includes vehicular density, RSU load, task size, and channel quality, and jointly optimises latency, energy consumption, task completion ratio, RSU utilisation rate, and handover overhead. Through extensive simulations, experimental results reveal that DRLO-VANET decreases the task execution latency by up to 40%, saves energy (30%–35%), and enhances the task completion ratio to over 90% with medium density, and decreases the handover frequency by nearly 50% analysis compared to the baseline methods. As a result, these results verify that policies activated by DRL outperform other energy management strategies by balancing short-term responsiveness with long-term system stability. We proposed a scalable, adaptive, and efficient framework with real-time task offloading capabilities for vehicular tasks. Its performance was proven through initial experiments and simulations, making it suitable for new autonomous transport systems that require high reliability and responsiveness.