This post is also available in:
עברית (Hebrew)
When a fire alarm sounds, most people instinctively head toward the nearest exit. The problem is that the closest route is not always the safest one. Fires can spread rapidly, smoke conditions can change within minutes, and an exit that appears safe when evacuation begins may become hazardous before occupants reach it.
Researchers have developed an AI-based evacuation system designed to address that challenge by predicting how a fire will evolve and continuously updating evacuation guidance in real time. The system, called Safe Step, combines machine learning, building sensor data, and fire modeling to identify safer escape routes as conditions change.
The technology is intended for use in smart buildings equipped with environmental sensors that monitor factors such as temperature, smoke, and air quality. The system works alongside dynamic emergency exit displays, electronic signs capable of changing directions during an emergency rather than simply pointing occupants toward the nearest exit.
According to TechXplore, unlike traditional evacuation algorithms, which focus on current conditions and shortest travel paths, it attempts to anticipate future hazards. The system uses reinforcement learning, a form of AI that learns through repeated trial-and-error simulations. During training, it analyzes building layouts and fire development scenarios generated using fire simulation tools, learning how different evacuation decisions affect occupant safety over time.
A key metric used by the system is the Fractional Effective Dose (FED), which measures cumulative exposure to toxic fire gases. Rather than simply calculating the fastest route, the AI seeks the path that minimizes total hazard exposure during evacuation.
For example, a conventional system might direct occupants toward the nearest exit even if smoke conditions are expected to worsen along the route. The system can forecast how the fire is likely to spread and instead direct people toward a farther exit if it offers lower overall exposure to toxic gases.
From a defense and public safety perspective, adaptive evacuation technologies could support emergency response in military facilities, critical infrastructure, transportation hubs, and other complex environments where large numbers of people must be moved safely during rapidly changing incidents.
Researchers are now working to expand the system beyond single-floor buildings. Future versions are expected to support multi-story structures and coordinate multiple occupants simultaneously, potentially reducing congestion while helping firefighters gain safer access to affected areas.
The research was published here.
