When shortest is not safest: Multi-agent evacuation with awareness and agile routing in dynamic hazards

Indoor environments can present significant spatial structural complexity, making safe and timely evacuation a critical challenge in the presence of dynamically evolving hazards and incomplete or delayed information available to evacuees. In this context, we study pedestrian evacuation in complex indoor environments using agent-based simulation, focusing on the performance of dynamic routing strategies under varying levels of situational awareness. We consider scenarios in which ambient intelligence systems detect pedestrians and hazards and provide real-time guidance through variable message panels and mobile applications. Using the open-source JuPedSim simulator extended with a customized wayfinding model, we evaluate four guidance configurations that combine low and high situational awareness with shortest-path and agile routing strategies. Low-awareness conditions represent locally informed evacuations, in which pedestrians rely only on reactive, locally available hazard information, while high-awareness scenarios provide dynamic risk updates during the evacuation process. Agile routing prioritizes structurally flexible paths, enabling adaptive rerouting when hazards evolve. Evacuation performance is assessed in terms of evacuation time and remaining-path risk (RPR), computed from the time-dependent node-level risk along the selected remaining route. Simulation results demonstrate that high situational awareness consistently improves both evacuation time and safety. Moreover, agile routing proves particularly effective when hazards emerge late or when spatial layouts are complex, especially under high-awareness guidance that supports adaptive route selection.