Scholarlink

<p>Unmanned Aerial Vehicle (UAV) assisted Mobile Edge Computing (UAV-MEC), with its high flexibility and mobility, enables efficient computing services for terminal devices (TD) in remote or emergency scenarios, with broad application potential. However, current UAV-MEC systems primarily employ binary offloading or basic dual-node partial offloading mechanisms. Although these simplify offloading decisions, they fail to fully exploit heterogeneous resources in dynamic and complex environments. To address this limitation, we propose a collaborative offloading framework based on Deep Reinforcement Learning (DRL), termed the Dual-Agent SAC-Based UAV Trajectory and Task Offloading Optimization (SAC-UTO). This framework incorporates Ground-MEC as an auxiliary computing node, forming a heterogeneous resource pool with UAV-MEC and local devices. The core innovation of SAC-UTO is its dual-agent system with distinct yet complementary roles: Agent 1 manages global strategy to optimize TD task offloading priorities, while Agent 2 dynamically adjusts UAV flight trajectory and task distribution across heterogeneous nodes based on these priorities. By incorporating flight distance and the impact of complex environments into the reward function for system latency, SAC-UTO effectively reduces UAV flight distance while preserving equivalent system latency.</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

Integrated task offloading scheduling and trajectory optimization for UAV-MEC using SAC-UTO

  • 期刊论文

  • 发表日期：

    2025年9月3日

查看全文
  • Kai Pan,
  • Jiadong Dong,
  • Chunxiang Zheng,
  • Xiaoxiao Wang

摘要

Unmanned Aerial Vehicle (UAV) assisted Mobile Edge Computing (UAV-MEC), with its high flexibility and mobility, enables efficient computing services for terminal devices (TD) in remote or emergency scenarios, with broad application potential. However, current UAV-MEC systems primarily employ binary offloading or basic dual-node partial offloading mechanisms. Although these simplify offloading decisions, they fail to fully exploit heterogeneous resources in dynamic and complex environments. To address this limitation, we propose a collaborative offloading framework based on Deep Reinforcement Learning (DRL), termed the Dual-Agent SAC-Based UAV Trajectory and Task Offloading Optimization (SAC-UTO). This framework incorporates Ground-MEC as an auxiliary computing node, forming a heterogeneous resource pool with UAV-MEC and local devices. The core innovation of SAC-UTO is its dual-agent system with distinct yet complementary roles: Agent 1 manages global strategy to optimize TD task offloading priorities, while Agent 2 dynamically adjusts UAV flight trajectory and task distribution across heterogeneous nodes based on these priorities. By incorporating flight distance and the impact of complex environments into the reward function for system latency, SAC-UTO effectively reduces UAV flight distance while preserving equivalent system latency.