Applying multi-agent reinforcement learning for escape path planning in a fire evacuation environment

Tools

Fitkau, Isabelle and Hartmann, Timo; Moreno-Rangel, Alejandro and Kumar, Bimal, eds. (2025) Applying multi-agent reinforcement learning for escape path planning in a fire evacuation environment. In: EG-ICE 2025. University of Strathclyde Publishing, GBR. ISBN 9781914241826 (https://doi.org/10.17868/strath.00093292)

[thumbnail of Fitkau-Hartmann-EG-ICE-2025-Applying-multi-agent-reinforcement-learning-for-escape-path-planning]
Preview
 Text. Filename: Fitkau-Hartmann-EG-ICE-2025-Applying-multi-agent-reinforcement-learning-for-escape-path-planning.pdf
Final Published Version
License: Creative Commons Attribution 4.0 logo
Download (2MB)| Preview

Abstract

Emergency evacuation planning is critical for building safety, yet current approaches often require costly re-modeling after design completion. This research addresses the gap between static evacuation simulations and dynamic environment optimization by implementing a multi-agent reinforcement learning framework where agents simultaneously learn escape paths and modify building components. Our system features a navigator agent seeking fast and efficient path trajectories and a door controller agent dynamically adjusting door positions across rooms. Using Proximal Policy Optimization, we tested three terminal configurations in a minimal MARL environment model. Results demonstrate successful agent cooperation without explicit communication, reducing episode lengths by 42-73% and developing door positioning strategies tailored to spatial constraints. These findings establish a foundation for future research on more complex evacuation scenarios, potentially contributing to improved building design and emergency response planning.

  • Item type:Book Section
    ID code:93292
    Dates:
    Date
    Event
    1 July 2025
    Published
    9 June 2025
    Accepted
    Subjects:Fine Arts > Architecture
    Department:Faculty of Engineering > Architecture
    Depositing user: Pure Administrator
    Date deposited:27 Jun 2025 15:12
    Last modified:01 Jul 2025 11:40
    URI:https://strathprints.strath.ac.uk/id/eprint/93292
CORE (COnnecting REpositories)