AI-Driven Infrastructure Design for Evacuation Efficiency and Multi-Hazard Resilience under Uncertainty

Building evacuations during fires, floods, earthquakes, and other disruptions depend on more than the physical layout of a building. They also depend on how hazards evolve, how building systems such as access control, notification, lighting, and smoke control perform, how occupants behave under stress, and how life-safety codes shape design decisions. Current building design workflows often evaluate these factors separately, which makes it difficult to predict evacuation performance when hazards, degraded systems, and human behavior interact. This project will develop artificial intelligence and machine learning methods to support code-aware building design under multi-hazard evacuation uncertainty. The research will create a design-stage digital twin that links hazard simulation, building system performance and degradation, and occupant behavior in a common modeling framework. Virtual reality experiments will be used to learn occupant behavior patterns, such as route choice, hesitation, group movement, and response to degraded information. Machine learning surrogate models will then provide fast, uncertainty-aware estimates of evacuation performance across many possible design and retrofit options. These models will be combined with constrained optimization methods that account for life-safety codes, cost, feasibility, and operational constraints. The framework will generate ranked portfolios of design and retrofit alternatives with estimated safety, resilience, cost, and robustness tradeoffs. Candidate options will be stress-tested under compound disruption scenarios, such as fire with power loss or blocked egress with degraded notification. The project will advance AI/ML methods for learning, optimizing, and evaluating infrastructure design decisions when physical hazards, human behavior, building systems, and regulatory constraints interact. The expected impacts include improved decision support for engineers, facility owners, emergency managers, and authorities having jurisdiction; open-source tools and scenario resources for research and education; interdisciplinary student training; K-12 outreach using infrastructure-safety and virtual-reality activities; and practitioner workshops with public-safety and industry partners. The long-term goal is to help communities make more transparent, evidence-based decisions for safer and more resilient buildings. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.