Scholarlink

<p>In complex and dynamic fire scenarios within high-rise buildings, the ability to plan and dynamically adjust 3D safe escape paths is crucial for ensuring personnel safety. Traditional path planning methods typically fail to incorporate multi-modal information from dynamic environments and do not account for the actual 3D spatial structures within buildings. This oversight significantly impacts the success rate of evacuations. To address this issue, we propose a dynamic planning method for 3D safe escape paths that utilizes multi-modal disaster infor mation sensing and deep reinforcement learning. Our method first establishes a multi-modal disaster environment sensing model by integrating various data sources, including vision, temperature, humidity, smoke concentration, and toxic gas concentration, to assess hazardous factors in real time. Second, we develop a multi-agent reinforcement learning model tailored to disaster environments, and design a dynamic planning algorithm for 3D paths that leverages multi-agent collaboration to optimize path selection in real time, thus adapting to the changing disaster environment. Experimental results demonstrate that our method surpasses traditional approaches in the efficiency and safety of escape path planning, particularly in complex 3D fire disaster scenarios.</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 three-dimensional safe escape path dynamic planning method based on multi-modal fire information sensing and deep reinforcement learning

  • 期刊论文

  • 发表日期：

    2025年11月17日

查看全文
  • Luxiu Yin,
  • Yu Deng,
  • Chun Wang,
  • Yaping Chen,
  • Wei Liang,
  • Kuanching Li

摘要

In complex and dynamic fire scenarios within high-rise buildings, the ability to plan and dynamically adjust 3D safe escape paths is crucial for ensuring personnel safety. Traditional path planning methods typically fail to incorporate multi-modal information from dynamic environments and do not account for the actual 3D spatial structures within buildings. This oversight significantly impacts the success rate of evacuations. To address this issue, we propose a dynamic planning method for 3D safe escape paths that utilizes multi-modal disaster infor mation sensing and deep reinforcement learning. Our method first establishes a multi-modal disaster environment sensing model by integrating various data sources, including vision, temperature, humidity, smoke concentration, and toxic gas concentration, to assess hazardous factors in real time. Second, we develop a multi-agent reinforcement learning model tailored to disaster environments, and design a dynamic planning algorithm for 3D paths that leverages multi-agent collaboration to optimize path selection in real time, thus adapting to the changing disaster environment. Experimental results demonstrate that our method surpasses traditional approaches in the efficiency and safety of escape path planning, particularly in complex 3D fire disaster scenarios.