Emergency egress risk assessment in high-rise residential buildings: predicting and assessing the behavior of people in fire emergency using simulation-powered predictive analytics, AI and VR
Yehuda Kalay, Professor
Architecture
Applications for Spring 2026 are closed for this project.
High-rise buildings pose unique fire-safety challenges due to high occupancy rate and difficult evacuation conditions. In 2023, San Francisco adopted a Fire Code amendment requiring full sprinkler retrofits in all residential high-rise buildings built before 1975, affecting 146 buildings, home to approximately 15,000 residents.
While sprinklers are an effective life-safety measure, retrofitting older, occupied buildings can be extraordinarily costly due to structural constraints and invasive construction practices. These costs will be passed on to residents, disproportionately impacting seniors and fixed-income households which may, in some cases, force them out of their homes and seek alternative accommodations in a city already suffering a severe housing shortage.
This mandate, therefore, creates a moral dilemma: while it is intended to protect life, it may instead make housing unaffordable.
We are developing a human behavior fire egress performance assessment for tall residential building, seeking a more nuanced approach that could achieve similar safety outcomes without displacing residents.
Our approach is centered on analyzing human behavior in high-rise buildings during fire and smoke conditions, using Agent-Based simulation, Predictive Analytics, Artificial Intelligence, and Virtual Reality tools. It concerns three main variables:
1. Space (physical characteristics of the built environment)
2. Actors (residents, firefighters, pets)
3. Activities (fire, smoke, egress behavior)
Role: We are looking for 2-3 URAP students who will help develop the simulation engine and the fire egress performance assessment.
Students are expected to be engaged in:
• Collaborative coding, troubleshooting, and applying various AI and other models to simulate the behavior of people in emergency conditions.
• Visualizing and reporting results.
• Reading, understanding, summarizing, and extracting key actionable insights and models from leading-edge papers in the field of fire safety, agent based simulation, and virtual reality.
• Be self-motivated, able to suggest development directions as well as critique such directions, and learn new methods as needed.
The research is conducted off-site, using weekly Zoom meetings, lasting 1-2 hours.
Qualifications: Applicants should possess skills in at least one of the following, and strong interest in the others:
• Knowledge of Search, Graph, Reinforcement Learning and other machine learning models (related courses such as CS188, CS189, CS186).
• Experience with agent-based simulation techniques.
• Experience with coding, especially related to applying machine learning models (preferably in Python).
• Interest and motivation to read research papers in a variety of fields describing AI / ML models.
• Data visualization and reporting (e.g. Pandas, Matplotlib).
Day-to-day supervisor for this project: Assistant Professor Hadas Sopher
Hours: to be negotiated
Off-Campus Research Site: Weekly zoom meetings
Engineering, Design & Technologies Digital Humanities and Data Science