Details published in the Journal of Building Engineering reveal a new approach to fire evacuation, moving beyond simply identifying the closest exit. Researchers at the National Institute of Standards and Technology (NIST) and colleagues have developed “Safe Step,” an AI model that forecasts fire evolution to dynamically update safe evacuation routes within a building. Unlike previous algorithms focused on shortest paths, Safe Step accounts for changing conditions, redirecting occupants away from exits that may become unsafe as a fire spreads. “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.” Some “smart” buildings are currently testing the technology with dynamic emergency exit displays utilizing real-time data from temperature and air quality sensors.
Safe Step: AI for Predicting Fire Spread & Evacuation
Safe Step, a newly developed artificial intelligence model from NIST, fundamentally alters conventional fire evacuation strategies by proactively forecasting hazard evolution, a capability previously absent in building safety systems. Unlike algorithms prioritizing shortest paths, this model doesn’t simply react to current conditions; it anticipates how a fire will spread, dynamically adjusting recommended escape routes to maximize occupant safety. Described in the Journal of Building Engineering, Safe Step utilizes reinforcement learning, a technique where the AI learns through simulated trial and error, leveraging data from NIST’s established fire simulation tools. The core of Safe Step’s effectiveness lies in its use of the fractional effective dose (FED) of toxic gases, a metric quantifying hazard severity over time; the model consistently selects routes minimizing FED exposure.
Researchers demonstrated the model’s superiority over traditional algorithms in test cases, particularly highlighting its ability to prevent occupants from unknowingly heading towards exits that become compromised as the fire grows. Future iterations envision an AI system with each instance representing an individual occupant, allowing the model to account for congestion and coordinate access for firefighters. Wai Cheong Tam, a NIST mechanical engineer, explained the team’s motivation: “We asked ourselves, ‘Can we build a better algorithm that predicts how the fire evolves, and in a way that helps save more lives?’” While widespread implementation remains five to ten years away, pending regulatory approval and rigorous testing, Safe Step represents a significant advancement toward intelligent firefighting and proactive building safety.
Reinforcement Learning Guides Safe Evacuation Route Selection
Current evacuation algorithms often prioritize the shortest path, a strategy proven insufficient when fires dynamically alter building safety. Researchers are now leveraging artificial intelligence to predict fire evolution and guide occupants toward truly safe exits. Unlike systems relying solely on current sensor readings, Safe Step forecasts how a fire will spread, allowing for proactive rerouting of evacuees. In comparative tests against traditional algorithms, Safe Step consistently identified safer evacuation routes, even in complex single-level structures. For instance, the model can recognize that an initially close exit might become compromised as a fire grows, and instead direct individuals to a more distant, but ultimately safer, alternative.
Fractional Effective Dose (FED) Metric Minimizes Hazard Exposure
Researchers are increasingly focused on quantifying hazard exposure during fire evacuations, moving beyond simple distance-based algorithms. This variable allows the “Safe Step” AI to prioritize routes minimizing cumulative hazard, rather than solely focusing on the shortest path to an exit. The lower the FED, the lower the risk for building occupants, providing a more nuanced evaluation of safety. Unlike previous approaches, Safe Step doesn’t merely react to current conditions; it actively forecasts how toxic gas concentrations will change as an evacuee traverses a route. This predictive capability is essential because a seemingly clear exit can quickly become dangerous as a fire evolves. Researchers demonstrated this with a test case involving a fire spreading into a hallway; while a traditional algorithm would direct someone to the closest exit, Safe Step could anticipate the increasing hazard and redirect them to a more distant, yet ultimately safer, alternative. The implementation of FED within the model requires numbers to determine the optimal route, and the algorithm’s effectiveness was validated through comparative tests against traditional methods using both simple and complex building structures.
“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,” said Fang.
Dynamic Exit Displays & Multi-Agent System Future Development
The integration of artificial intelligence into building safety systems is moving beyond theoretical models and into practical testing, with dynamic emergency exit displays now being piloted in several “smart” buildings. These displays utilize real-time data from sensors monitoring temperature and air quality to assess exit safety, offering occupants guidance beyond simply the nearest route. This proactive approach represents a shift from reactive emergency signage, and builds upon the “Safe Step” AI model developed by NIST researchers, detailed in the Journal of Building Engineering. Researchers are already looking beyond single-story applications, with ongoing work focused on adapting the model to handle the complexities of multi-level structures. To more accurately simulate real-world scenarios, the team intends to create an AI system employing multiple agents, where each agent represents an individual building occupant.
Interactions between these agents will allow the model to account for congestion, such as bottlenecks at building entrances, and dynamically adjust evacuation routes accordingly. This advanced coordination isn’t limited to occupant safety; the model could also facilitate firefighter access. By directing evacuees to alternative exits, the system could open pathways for emergency responders, improving their ability to extinguish fires and assist vulnerable individuals. NIST estimates that technologies like Safe Step could begin appearing in buildings within five to ten years, though widespread implementation hinges on regulatory approval and thorough reliability testing. “This research is still in the early research and development stage, but it represents an important step toward intelligent firefighting where effective use of advanced technologies can protect property and save lives,” Fang concluded, emphasizing the potential for AI to fundamentally improve fire safety protocols.
“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.”
