Being a firefighter in a blaze is an immensely disorienting experience. Your sight is obscured by smoke and flames, your hearing overcome with the sound of crackling fire and wood. You don’t know the layout of the building you’re entering, or its potential dangers. Is there a propane tank in the basement? Is there a child needing rescue in a smoke-veiled back bedroom?
An artificial intelligence system developed by NASA’s Jet Propulsion Laboratory aims to make this hazardous experience safer. AUDREY (Assistant for Understanding Data through Reasoning, Extraction, and sYnthesis) can guide firefighters by tracking their movements, identifying hazards and facilitating communication.
“Back in 1999, firefighters got killed in a building because they were trapped and they couldn’t find their way out,” says Edward Chow, manager of the Jet Propulsion Laboratory’s Civil Program Office and program manager for AUDREY. He refers to the fire at the Worcester Cold Storage warehouse in Worcester, Massachusetts, in which six firefighters died after becoming lost in the windowless, maze-like former meat storage facility. “[In a fire] you can’t really see anything, you lose orientation. What we want to do is apply AUDREY technology to help firefighters get to a safe place,” he adds.
The firefighter project is a collaboration between the Jet Propulsion Laboratory and the Department of Homeland Security, but the technology stems from the lab’s work on space rovers. The rovers used on Mars, Chow explains, are programmed with maps of the planet’s surface. They know when to expect a rock and when to expect a valley. This is possible because NASA has extensive information about Mars’ surface. But now that NASA is exploring lesser-known terrain such as Europa, Jupiter’s largest moon, a more advanced technology is needed.
Europa, Chow explains, is covered in ice, which is believed to conceal an ocean.
“There’s no way for us to map out what’s under the ice because you can’t see it,” he says. “So when we send a spacecraft to Europa and it’s swimming around, we really have no idea what we’ll be faced with. So we need to make this system a lot smarter.”
This means creating an artificial intelligence that works much more like a human brain. Rather than telling it to expect a rock, the system should be able to understand the concept of ‘rock’ and map it when it sees one.
Fires, Chow says, are more like being on Europa than being on Mars.
“Every fire firefighters respond to is different from the previous fire,” he says. “No two fires are identical. So traditional artificial intelligence won’t work.”
The technology being developed for the Europa mission, and the technology behind AUDREY, is about creating this sort of human-like intelligence to respond to unknown situations. The AI begins with a small amount of human input—a “bootstrap,” Chow calls it—and builds on it through experience. In the case of AUDREY’s firefighting application, these bootstraps might involve an experienced firefighter teaching the AI things like which flame colors correspond to the hottest temperatures, or what it means when you see smoke but no flames. The AI then learns more by watching real-life scenarios.
In the field, firefighters wear sensors, which communicate with each other and with AUDREY in the cloud. The cloud-based AUDREY can create a map of the terrain and give firefighters on the ground warnings and directions. Chow’s team also hopes to make it possible for AUDREY to communicate with Internet of Things objects such as thermostats already present in the location to gain more information.
The AUDREY team has already worked with first responders. Over the next eight months, they plan to do field trials involving firefighters. After that, they hope to transfer the technology to private industry for development. They estimate AUDREY might be available for use within the next several years.
For American firefighters—68 of whom died while on duty last year—this is artificial intelligence with a very human purpose.