Dynamic exit displays go beyond static signs as a new AI model, Safe Step, provides real-time guidance during a building fire. Developed by researchers at the National Institute of Standards and Technology (NIST) and detailed in the Journal of Building Engineering, Safe Step does more than simply react to current conditions. Predict the spread of fire and proactively adjust evacuation routes. “Fires can grow and spread,” said Hongqiang “Rory” Fang, a researcher at NIST and lead author of the paper. “Our model predicts how a fire will progress and helps update exit signs to direct people to the safest exit.” Unlike traditional algorithms that prioritize the shortest path, Safe Step uses reinforcement learning to minimize exposure to hazards and calculate a fire safety metric called the partial effective dose of toxic gases to guide occupants along safer routes.
Safe Step: Fire spread prediction and evacuation using AI
Unlike previous systems that rely on identifying the shortest path to exit, Safe Step uses reinforcement learning to assess cumulative risk and takes into account the changing toxicity of the gas over time. The researchers employed a metric called fractional effective dose (FED) to quantify hazardous exposure and used a model that prioritized routes with the lowest FED values. Testing against traditional algorithms on both simple and complex single-level building structures demonstrated Safe Step’s consistent ability to identify safer evacuation routes. For example, AI can guide occupants away from nearby visible exits if it predicts that exits will be at risk as the fire spreads. Although currently being validated on a single-story layout, the team is now focused on extending Safe Step’s functionality to multi-story buildings and incorporating behavior models for large numbers of evacuees at the same time. This advancement has the potential to address exit congestion and even coordinate firefighter access to facilitate rescue and firefighting operations.
Reinforcement learning guides safe evacuation route selection
Building safety systems are becoming increasingly sophisticated, moving beyond static signage with new approaches that leverage artificial intelligence to guide occupants during emergencies. Current systems often rely on directing individuals to the nearest exit, but these methods fail to account for the dynamic nature of fire spread and the resulting hazards along potential escape routes. The model, published in the Journal of Building Engineering, uses reinforcement learning, a technique in which AI learns through trial and error, to predict fire progression and optimize evacuation strategies. In testing, Safe Step demonstrated its ability to identify safer alternatives even when the nearest exit is compromised. For example, move residents away from smoke-filled hallways and direct them to safer, more remote routes. Future iterations envision an AI system in which each occupant is represented as an individual to better model crowding and coordinate firefighter access. Fang said, “Although this work is still in the early stages of research and development, it is an important step toward intelligent firefighting that can protect property and save lives through the effective use of advanced technology.”
Minimize hazardous exposure with fractional effective dose (FED) metrics
This nuanced approach allows the AI to take into account changing conditions and predict how the concentration of toxic gases will change as individuals pass through a building during a fire. The implementation of FED allows Safe Step to make more informed decisions rather than simply reacting to current conditions. For example, the model can evaluate scenarios where an initially obvious exit becomes dangerous as the fire spreads, directing occupants to a more distant but ultimately safer alternative exit. “We asked ourselves, ‘Can we build better algorithms in a way that predicts how a fire will progress and saves more lives?'” This predictive ability is critical because the model does not require real-time fire simulations, said NIST mechanical engineer Wai Cheong Tam. Instead, it relies on live sensor data to continually refine its recommendations. During tests on complex single-story building structures, Safe Step consistently identified safe evacuation routes and outperformed traditional algorithms in scenarios where conditions rapidly deteriorate.
Future development of dynamic exit display and multi-agent system
These signs are not static signs. You can now indicate whether an exit is safe or direct residents to a different path. This feature is being tested in several “smart” buildings where the technology is currently being deployed. Important future developments include moving from single-agent models to multi-agent systems that calculate the optimal route for a single evacuee. This approach envisions each building occupant being represented as an individual “agent” within the AI, allowing for a more nuanced understanding of crowd dynamics and potential bottlenecks. Wai Cheong Tam pointed to the possibility of a coordinated response, saying the team is looking at ways to direct evacuees to different exits in this model while coordinating access points for firefighters at the same time. Such systems could reduce exit congestion, a common hazard during emergencies, and facilitate more effective firefighting and rescue operations, especially for vulnerable populations. Researchers predict that technology like Safe Step will begin to be deployed in buildings within five to 10 years, subject to regulatory approval and rigorous reliability testing, ultimately representing an important step toward intelligent fire suppression and enhanced life safety.
“Fires can grow and spread,” said Hongqiang “Rory” Fang, a researcher at NIST and lead author of the journal paper. “Our model predicts how a fire will progress and helps us update emergency exit signs to direct people to the safest exit.”
