Scholarlink

<p>Efficient collaboration among unmanned vehicles (UVs), such as unmanned aerial vehicles (UAVs) and unmanned ground vehicles (UGVs), through mobile crowdsourcing (MC) in ubiquitous computing applications remains a challenging issue. Leveraging mobile sensing signals can enhance versatility and efficiency. For example, tasks such as pathfinding may require cooperation among UAVs, UGVs, and sensors. Addressing collaborative challenges, however, often involves managing heterogeneity in aspects such as velocity, sensing range, and complex communication coordination among UVs. In this study, we propose the Heterogeneous Graph Neighborhood-based Multi-Agent Asynchronous Edge Reinforcement Learning (HGN-MAAERL) framework to analyze both individual and cooperative interactions between UAVs and UGVs. By optimizing factors such as offloading latency, data collection efficiency, cooperation levels, reliability, and energy consumption, HGN-MAAERL enables asynchronous agents to facilitate real-time data collection in dynamic edge environments. Specifically, the Heterogeneous Mask State-assisted Transformer (HMST) serves as the policy and value network architecture, supporting temporal modeling for heterogeneous crowdsourcing. Additionally, the Individual and Cooperation Graph Neighborhood (ICGN) module captures external, individual, and cooperative advantage functions through a Bayesian optimization procedure to model cooperation preferences between UAVs and UGVs. Empirical evaluations demonstrate that HGN-MAAERL outperforms other frameworks in terms of collaboration effectiveness, especially when assessed on NCSU and Purdue MC benchmarks for sensor interaction and sensing range coordination.</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

HGN-MAAERL: Hetergeneous graph neighborhood-based multi-agent asynchronous edge reinforcement learning for efficient multiple unmanned vehicles collaboration

  • 期刊论文

  • 发表日期：

    2025年9月15日

查看全文
  • Feiyang Huang,
  • Ruihan Hu,
  • Qiming Cao,
  • Xiangdong He

摘要

Efficient collaboration among unmanned vehicles (UVs), such as unmanned aerial vehicles (UAVs) and unmanned ground vehicles (UGVs), through mobile crowdsourcing (MC) in ubiquitous computing applications remains a challenging issue. Leveraging mobile sensing signals can enhance versatility and efficiency. For example, tasks such as pathfinding may require cooperation among UAVs, UGVs, and sensors. Addressing collaborative challenges, however, often involves managing heterogeneity in aspects such as velocity, sensing range, and complex communication coordination among UVs. In this study, we propose the Heterogeneous Graph Neighborhood-based Multi-Agent Asynchronous Edge Reinforcement Learning (HGN-MAAERL) framework to analyze both individual and cooperative interactions between UAVs and UGVs. By optimizing factors such as offloading latency, data collection efficiency, cooperation levels, reliability, and energy consumption, HGN-MAAERL enables asynchronous agents to facilitate real-time data collection in dynamic edge environments. Specifically, the Heterogeneous Mask State-assisted Transformer (HMST) serves as the policy and value network architecture, supporting temporal modeling for heterogeneous crowdsourcing. Additionally, the Individual and Cooperation Graph Neighborhood (ICGN) module captures external, individual, and cooperative advantage functions through a Bayesian optimization procedure to model cooperation preferences between UAVs and UGVs. Empirical evaluations demonstrate that HGN-MAAERL outperforms other frameworks in terms of collaboration effectiveness, especially when assessed on NCSU and Purdue MC benchmarks for sensor interaction and sensing range coordination.