Teaching Drones to Sniff Out Toxic Air
Swarms of the flying devices, using sensors and AI, will learn to find and track harmful gases
Not a week goes by, it seems, without more news of how drones are going to make our lives so much easier or what they can do now to entertain us. Most recently, there were reports of the flying devices delivering food to golfers on a course in North Dakota and being used as backup dancers at Drake’s shows.
But far away from the back nine and concert stages, autonomous flying vehicles are doing serious business, from helping to save lives during hurricanes to lending aging farmers a hand. And, if a Rice University research project comes to fruition, a swarm of drones could one day work together to sense toxic gases in the atmosphere and map out a safe perimeter.
Boosted by a recent $1.5 million National Science Foundation grant, the scientists, in collaboration with the Baylor College of Medicine and Technology For All, a Houston nonprofit, will focus on giving drones the intelligence to sniff out where dangerous pollution has spread following explosions or leaks, particularly after extreme weather events.
In fact, noted Edward Knightly, the electrical and computer engineer heading up the research, that kind of near disaster happened not far away following Hurricane Harvey last year when first responders were sickened by a gas leak from a chemical plant near Houston.
“The first responders went in to set up a perimeter of about two and a half kilometers in order to warn neighborhood residents,” he says. “But those first responders were themselves hospitalized. The perimeter was too close to the source, but they didn’t know that. Nobody knew.
“And, that’s a great example of where we could send in drones from a safe distance and they can figure out where’s a safe boundary, the effect of the wind, whether there have been chemical reactions in the air. Finding and setting the right boundaries is a complex decision, and we think drones could be a big help in making that decision.”
But giving drones the ability to identify volatile gases in real-time brings some big challenges. For starters, can effective sensors be scaled down to a size and weight able to fit on a drone? The team is working with Frank Tittel, an emeritus professor at Rice and a pioneer in the use of lasers to detect the signatures of molecules.
“We have to shrink the sensors down to one and a half kilograms or less,” Knightly says. “He’s developed ways for lasers to sense gas concentrations for a wide variety of gases. But typically, he designs systems to be used on the ground. He’s never had to deal with creating sensors with this mass or size constraint.”
Much of the project’s focus to this point has been to sharpen the intelligence of the drone fleet, known as ASTRO. While the focus going forward will be on gas-sensing capabilities, ASTRO’s learning has been broader in scope. The goal has been to develop a system where the drones can work together—without human intervention—to find objects.
Knightly explains that the team has trained its fleet of four drones to track down a wireless device, including one on a moving golf cart. And, it’s done “off-the-grid,” without a human on the ground directing things with a remote control.
“We want the network of drones not only to be able to fly amongst themselves, but even if there is no air-to-ground communication available,” he says. “If there’s extreme weather and infrastructure is down, we still want to be able to send off drones to collect information about a hazardous area.”
Before they begin to track anything, the drones use their artificial intelligence to learn about their surroundings. It’s what Knightly refers to as the “search and learn phase.” The drones start missions by first spreading out to literally get a lay of the land, all the time communicating with each other so they don’t bunch up in one area.
“Before they can locate and track a wireless device, for instance, it’s important for them to learn the environment,” says Knightly. “Are there a lot of trees? Tall buildings? Or is it wide open spaces? The goal is to learn the environment so they can accurately track.”
Swarm and track
Then they move on to phase two, which is aptly called “swarm and track.” Again, when the target object is a wireless device, they’ve identified where its signal is the strongest, and that’s where the drones come together as a swarm. In the case of a hazardous event, it could be where a gas reading is the strongest.
“They determine that this is what we should be measuring, so let’s go collect some high-resolution data,” says Knightly.
“Of course, gases all have their own spectral signatures,” he adds. “When the drones go out, there’s going to be a mix of different gases. It’s not going to be a clear signal of just one. So we need the drones to learn about the environment, compare it to statistical baseline models we’ve developed, and then be able to identify the sources of hazardous emissions and the boundaries of where they’ve spread.”
Luca Mottola, an associate professor in the Department of Electronics, Information and Bioengineering at the Polytechnic University of Milan, believes this is a "paradigmatic example" of when drones can be more effective than any other technology.
"The ability to roam unconstrained, while carrying rich sensor payloads, gives them a unique set of abilities," says Mottola, a leading expert on wireless and drone sensor networks. "Currently available technology to address similar scenarios heavily relies on human intervention and is therefore way slower and prone to errors.
"So, the impact of deploying drone technology is therefore manyfold. Responding more quickly may provide improved protection of assets, limit damage to structures and ameliorate the impact on human lives."
The Rice researchers hope to eventually test their gas-sensing drones in an ideal location nearby—the Houston Ship Channel, with its numerous chemical refineries and industrial processing plants. As the project proceeds, the fleet could expand to as many as ten drones.
The team has also developed a mobile app with the idea that it could use the drone sensor findings to send real-time air quality warnings to residents’ phones.
“We’d bring in doctors to help set thresholds for a community,” Knightly says. “When is it a typical day? And when is it a day to shelter in place?”