NIST builds an AI model that picks the least toxic escape route during a building fire
NIST researchers have developed Safe Step, a reinforcement learning model that selects evacuation routes minimising exposure to toxic gases, with commercial use estimated at five to 10 years away.
Researchers at the U.S. National Institute of Standards and Technology have developed an artificial intelligence model called Safe Step that recommends the safest evacuation route during a building fire, according to a paper published online on January 15, 2026, in the Journal of Building Engineering and reported by the agency on June 4 (DOI 10.1016/j.jobe.2025.115132).
The model uses reinforcement learning, finding safe paths through trial and error. It is trained using building floor plans together with data from NIST fire simulation tools to anticipate how a fire will develop, NIST said. In use, it would not run simulations in real time but would update its recommendations from building sensor data. Route choice is driven by the fractional effective dose metric for toxic gases, so the model selects the path with the least exposure — a departure from existing algorithms, which NIST said consider only the current state and cannot account for cumulative risk.
Safe Step is designed to work with electronic dynamic exit signs that can show whether an exit is usable or display directional arrows. The researchers validated it against existing algorithms in two test cases and in a more complex single-storey building layout. The current model works only on single-floor plans.
NIST said the team plans to extend the model to multi-storey buildings and to build a multi-agent AI system that assigns an agent to each occupant, allowing crowding at doorways to be spread out and firefighter access routes to be coordinated. The agency estimated commercial availability at five to 10 years away, saying uptake would hinge on how the technology performs in reliability testing, whether regulators sign off on it and how well it can be tied into the safety systems already in place. NIST has conducted fire safety research for more than a century, and says improvements to smoke alarms and firefighting equipment have contributed to falling fire deaths.
The paper was written by Michael Xuelin Huang, Md. Ismail Siddiqi Emon, Ruggiero Lovreglio, Wai Cheong Tam and Hongqiang Fang.