When does helping reduce evacuation efficiency? Modeling interaction behaviors and their trade-offs in multi-floor built environments

Abstract

Emergencies in multi-floor buildings expose more than architectural bottlenecks; they reveal the trade-offs embedded in human behavior. This study presents an agent-based model that simulates queuing, overtaking, competitive and helping behaviors, and the negative effects of physical interaction (e.g., falling, trampling, tripping and injury) during evacuation. Applied to a virtual four-floor building, 36 simulation experiments reveal four key findings. First, total evacuation time is dominated not by average collective behavior but by the final few agents who move slowly or travel long distances. Second, queuing behavior linked to shared social identity can reduce conflict but slows down 98.6 percent of agents. Third, shared social identity increases crowd stillness time and risk of falling. Finally, helping behaviors, especially toward elderly agents, may significantly delay evacuation when combined with spatial disadvantage. These results highlight a critical tension between social cohesion and systemic efficiency, suggesting the need for behavior-aware evacuation strategies.