A team of researchers at the National Institute of Standards and Technology (NIST) has created an artificial intelligence (AI) model that is capable of redirecting occupants of a building to the safest evacuation route during a fire.

The device, dubbed Safe Step, can reportedly be used with electronic displays to show whether an exit is safe to use.

Source: A. Kim/NIST

"Fires can grow and spread," explained the NIST research team. "Our model forecasts how the fire is evolving and can help update emergency exit displays to direct people toward the safest exit."

According to its developers, the Safe Step can be used in "smart" buildings, where sensors monitor real-time environmental conditions — like temperature and air quality.

The Safe Step uses reinforcement learning — a type of AI — to determine the safest routes through a building via trial and error. To accomplish this, Safe Step relies on the layout of the building to learn evacuation routes along with data from a NIST fire simulation tool to predict how a fire in the layout develops over time.

As the system trains, the model learns to make predictions about how a fire will impact occupants and subsequently guides them to safer evacuation routes. When used in a real-world setting, the model does not need to run a simulation of the fire in real time. Instead, it relies on live sensor data gathered from the building, which enables it to continuously adjust its recommendations as the fire grows.

However, the algorithm requires numbers to confirm whether it's selecting the optimal route. As such, the team relied on a fire safety metric: the fractional effective dose (FED) of toxic gases. The team explained that this variable demonstrates the severity of fire hazards to which a person is exposed over time. For instance, the lower the FED, the lower the hazard exposure for the occupants. The model selects the path with the lowest FED, taking into account how toxic gas exposure shifts over time as the occupant moves.

The team tested the Safe Step AI model against traditional evacuation algorithms and determined that it consistently identified safer escape routes, even in more complex building layouts. Unlike conventional systems that guide occupants to the nearest exit, Safe Step can forecast how a fire and smoke might spread over time and subsequently reroute occupants to safer exits in the event that conditions worsen.

For now, the new model is designed for single-story buildings, but future iterations will focus on multilevel structures and multi-agent systems that can account for the movements of many occupants. The team believes that this will help reduce congestion, coordinate evacuations with firefighter access and improve protection for vulnerable individuals.

An article detailing the work, “Development of a tenability-based path planning model for building fire evacuations using reinforcement learning,” appears in the Journal of Building Engineering.