Researchers at the National Institute of Standards and Technology (NIST) have developed an AI model designed to guide building occupants toward safer evacuation routes as fire conditions change.
The model, called Safe Step, can forecast how a fire may spread through a building and help dynamic emergency exit displays redirect occupants away from routes that could become unsafe. The work is described in the Journal of Building Engineering.
“Fires can grow and spread,” said Hongqiang “Rory” Fang, a research associate at NIST and first author of the journal paper. “Our model forecasts how the fire is evolving and can help update emergency exit displays to direct people toward the safest exit.”
Safe Step is intended for use in smart buildings equipped with sensors that monitor real-time conditions such as temperature and air quality. In those settings, electronic exit signs could display whether an exit remains safe or point occupants toward an alternative route.
According to NIST, previous evacuation route algorithms have focused on identifying the shortest safe path based on current conditions. However, those approaches do not account for the cumulative hazards people may encounter as they move through a building during a developing fire.
“We asked ourselves, ‘Can we build a better algorithm that predicts how the fire evolves, and in a way that helps save more lives?’” said NIST mechanical engineer Wai Cheong Tam.
Safe Step uses reinforcement learning, a form of artificial intelligence in which a model learns through trial and error. The system uses a building’s layout to learn possible evacuation routes and draws on data from a NIST fire simulation tool to anticipate how fire conditions may develop over time.
During training, the model learns to predict how a fire could affect occupants and then identifies routes that reduce exposure to hazardous conditions. In a real building, the model would not need to run a full fire simulation in real time. Instead, it would use live sensor data to update evacuation guidance as conditions change.
To determine the safest route, researchers used a fire safety metric known as fractional effective dose, or FED, which measures the severity of exposure to toxic gases over time. A lower FED indicates lower hazard exposure, allowing Safe Step to prioritize the route that minimizes risk as an occupant moves through the building.
The researchers tested the model against a traditional path-planning algorithm in two scenarios and also applied it to a more complex single-level building layout. NIST said Safe Step consistently identified safer evacuation routes.
In one example, a conventional algorithm might direct an occupant across a hallway toward the closest exit. If smoke and toxic gases spread quickly, that route could become more dangerous by the time the person reaches it. Safe Step is designed to anticipate that change and direct the occupant toward a farther but safer exit.
The current version of Safe Step is designed for a single-story floor plan. Researchers are now working to expand the model for multilevel buildings, where occupants may need to move between floors as well as along corridors.
NIST also plans to develop a multi-agent version of the system, with each agent representing a different building occupant. That approach could help the model account for congestion, occupant movement and the interaction between evacuees during a real emergency.
For example, if a bottleneck develops at one entrance, an improved system could direct some evacuees to other exits while helping coordinate access routes for firefighters entering the building.
The researchers estimate that technologies such as Safe Step could begin appearing within five to 10 years, although adoption would depend on regulatory approval, reliability testing and integration with existing fire and life safety systems.
NIST said the research forms part of its wider work to advance fire safety through improved detection, protective equipment and data-driven emergency response technologies.
“This research is still in the early R&D stage, but it represents an important step toward intelligent firefighting where effective use of advanced technologies can protect property and save lives,” Fang said.
