Scholarlink

Achieving quick and safe evacuation in densely populated areas remains challenging because the preplanned routes and fixed signs are inadequate for responding to fast-moving dangers or crowd congestion. We propose a four-layered feedback system with live data on the mobile devices of people and sensors in buildings and augmented reality (AR) instructions and a deep reinforcement learning (RL)-operated optimizer. The RL agent transforms the dynamic environment into a multidimensional state tensor, and it produces customized egress routes that trade off travel duration, traffic, and vulnerability; the AR front end displays such guidelines in place on head-mounted or smartphone devices. The platform is prototyped on Rhinoceros 7 and is assessed on a parameter’s matrix using the run types of 1800 at 3 building types, 3 crowd densities, and 4 hazard setups calculated in 3 h. In contrast to regulatory minimum static signage, the improved SWPR system reduces the average evacuation time by 49% (312 s → 158 s), increases the safety factor from 0.78 to 0.92, and decreases peak people-over-area from 5.2 to 2.1 persons/m2, even though the route recalculation latency is below 65 ms. Such findings indicate that crowdsensed information and AR-based visual aids and on-device RL-based decision-making have the potential to assist in evacuation in future smart cities. This chapter combines building design, everywhere sensing, and elastic algorithms into a complete system and solid evidence of effective emergency management in dense urban territory.

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

Leveraging Augmented Reality and Reinforcement Learning for Optimizing Evacuation in Crowded Public Spaces

  • 章节

  • 发表日期：

    2026年

查看全文
  • Mohammad Anvar Adibhesami,
  • Maryam Alsadat Ziaei Mazinan,
  • Bonin Mahdavi Estalkhsari,
  • Bahare Mirzaei,
  • Mahyar Pajani

摘要

Achieving quick and safe evacuation in densely populated areas remains challenging because the preplanned routes and fixed signs are inadequate for responding to fast-moving dangers or crowd congestion. We propose a four-layered feedback system with live data on the mobile devices of people and sensors in buildings and augmented reality (AR) instructions and a deep reinforcement learning (RL)-operated optimizer. The RL agent transforms the dynamic environment into a multidimensional state tensor, and it produces customized egress routes that trade off travel duration, traffic, and vulnerability; the AR front end displays such guidelines in place on head-mounted or smartphone devices. The platform is prototyped on Rhinoceros 7 and is assessed on a parameter’s matrix using the run types of 1800 at 3 building types, 3 crowd densities, and 4 hazard setups calculated in 3 h. In contrast to regulatory minimum static signage, the improved SWPR system reduces the average evacuation time by 49% (312 s → 158 s), increases the safety factor from 0.78 to 0.92, and decreases peak people-over-area from 5.2 to 2.1 persons/m2, even though the route recalculation latency is below 65 ms. Such findings indicate that crowdsensed information and AR-based visual aids and on-device RL-based decision-making have the potential to assist in evacuation in future smart cities. This chapter combines building design, everywhere sensing, and elastic algorithms into a complete system and solid evidence of effective emergency management in dense urban territory.