Credit to Author: Douglas Starr| Date: Tue, 28 Feb 2017 10:45:14 +0000
Imagine you’re part of a great swelling crowd, one of 60,000 people who fill up the cauldron of noise and chaos that is a sold-out football stadium. For you and everyone around you, the game is an open-air gathering place, a chance to steam and scream and worry about nothing except the other team’s menacing D. To the security officials responsible for your safety, it is a constant source of worst-case-scenario planning. They install metal detectors; they enlist a kennel’s worth of bomb-sniffing dogs; they plant concrete pillars around the perimeter to keep out cars; they train personnel in the dark art of bag searching; they even obtain a temporary flight restriction from the FAA to keep all aircraft above 3,000 feet for a radius of 3 miles. They spend millions of dollars and thousands of hours to keep you safe, yet they know that none of it can stop a 3-pound off-the-shelf drone from flying in and dropping something on the crowd. Maybe it’s a toxic mist. Maybe it’s a bomb. Whatever it is, you’ll never see it coming, and because there is currently no legal way to bring down a drone with any accuracy or reliability, there’s nothing anyone can do but wait for it.
In the summer of 2015, Ross Lamm and Dave Romero watched just such a scenario unfold from within a skybox at a large university stadium. The head of security for the college, fearful of the damage drones could do, had decided to run a simulation of a drone attack inside his 60,000-capacity football stadium. (The university asked that identifying details be withheld so as not to share its playbook with would-be attackers.) Campus officials launched a DJI quadcopter, a midsize, midpriced drone, and steered it toward the bleachers, pretending to spread nerve gas on the hundred students gathered below. As the drone looped lazily over the crowd, some of them pretended to vomit convulsively, some twitched spasmodically, some staggered like zombies and then collapsed. Emergency personnel rushed in, assessing the pretend damage and carrying pretend victims out to vans equipped as medical stations.
Up in a skybox, Lamm and Romero, cofounders of Black Sage Technologies, monitored the drone-tracking equipment they’ve spent the past few years developing. Almost immediately after the drone lifted off, Lamm and Romero’s radar detected it. Their AI-powered software identified it as a drone (and not, say, a bird), and their tripod-mounted cameras tracked it as it made its way over the crowd. As they heard the ominous buzzing overhead and watched the college kids pretend to die, Romero and Lamm allowed themselves a small measure of satisfaction—Black Sage’s tracking system worked, and in the event of an actual attack it could give authorities a few crucial extra minutes to mobilize. Mostly, though, Romero and Lamm felt alarmed, knowing all they could do was watch. “Holy shit,” Romero remembers thinking. “We can do everything but stop this catastrophic incident from occurring.”
Shaken and stirred, they returned to Black Sage’s headquarters in Boise, Idaho, and spent a year enhancing their system so that it can now not only track drones but also bring them safely to the ground using radio-frequency-jamming technology. There is only one small hitch: Like almost every drone-interdiction technology in development, frequency jammers run afoul of several US laws, most of which were passed when people hadn’t dreamed of owning their own unmanned aircraft. Romero and Lamm’s solution to the mock terror in the stadium—a solution that they have shown can reliably counter the threats drones pose to targets as varied as prisons, airports, and arenas—is illegal here, which leaves the future of Black Sage’s technology, like the future of drones themselves, very much up in the air.
The two inventors met in 2013 through a mutual friend in Boise. Romero, 31, grew up on a 2,200-acre cattle ranch 50 miles south of the city, the prototypical boy-tinkerer making miracles out of scrap metal. He built lots of dune buggies, motorcycles, and other contraptions, most of which worked, one of which burst into flames. He taught himself computer programming on his family’s IBM 386. After graduating from college in 2007, he started a software company called Tsuvo that performed regression analysis—taking large data sets from disparate government agencies, some of which involved thousands of statistics, and distilling them into clean, color-coded graphics that even nonstatisticians could understand. This kind of massive data crunching and predictive analysis, useful to bureaucracies both here and abroad, led him to live for varying amounts of time in Chile, Palau, and finally, Thailand. It also introduced him to the power of machine-learning algorithms, which helped make quick work of even the thorniest data sets.
In the aftermath of al Qaeda’s attack on the USS Cole, Lamm helped develop a robotic vision system that allowed ships to detect and quickly respond to speedboat attacks.
Where Romero is an adrenaline fiend—ask about the mountain bike perched in his office and he’ll show you a photo of himself on the bike, halfway through a backflip—Lamm, 45, likes nothing more than sailing with his two sons on a quiet lake. He is deliberate and thoughtful, choosing his words carefully, not out of caution but from an engineer’s appreciation of what’s precise and what’s not. While earning a PhD concentrated on machine vision in the late ’90s, he developed an algorithm that enabled a tractor-mounted camera to tell the difference between cotton plants and weeds, allowing farmers to spray herbicide more accurately. In the aftermath of al Qaeda’s attack on the USS Cole in 2000 (an explosive-laden speedboat crashed into the ship, killing 17 sailors), he helped a US Navy and Coast Guard contractor develop a robotic vision system that allowed ships to detect and quickly respond to speedboat attacks. (With your own vessel rocking and an enemy boat closing in fast, it’s surprisingly difficult to track ships on the water.) He also took part in constructing the warning system in Washington, DC, that locks onto commercial airplanes that drift into restricted airspace and beams an unmistakable red-red-green, red-red-green laser signal into the cockpit to alert the plane’s pilots to fly elsewhere. After more than a decade living and working in Napa Valley, California, he relocated to Boise in 2012, in part so his wife could move her winery there.
Detect, Identify, and Defeat
Lamm and Romero first crossed paths when their mutual friend asked for their help landing a government contract: The state of Idaho wanted to install a new warning system on a highway to prevent cars from crashing into animals after dark. The existing warning system flashed a light whenever a deer or an elk crossed the road, but because the signal would also light up whenever the wind sent leaves and branches tumbling across the pavement—which was often—drivers came to ignore the warning lights altogether. The highway developed one of the highest wildlife crash rates in the state, and when Romero was home from Thailand for a month visiting his family for Christmas, the friend invited him and Lamm to a brainstorming session at a coffee shop. Could some combination of Lamm’s expertise in robotic vision and Romero’s experience with machine learning help solve the highway problem? “After our friend introduced us, he hardly got a word in,” Romero says. “We got into this virtuous cycle of building on each other’s ideas.”
The pair got to work. Near the highway, they set up a Doppler radar (to detect moving objects) along with an infrared camera (for nighttime viewing) and routed the output to Romero, who had returned to Thailand for a few months to finish some work. To train his machine-learning algorithms to distinguish between animals and clutter, he would spend 45 minutes of his lunchtime each day (perfect for nocturnal sightings in Idaho) watching the infrared images and signaling yes or no as to whether they were wildlife. The system accumulated thousands of data points on the moving objects that crossed the camera’s field of view—speed, acceleration, direction—and once that data was correlated with Romero’s yes/no designations, the algorithm learned to recognize what probably was an animal and what probably wasn’t.
“It’s a beautiful algorithm that takes data from radar and enriches it with close probabilities,” Romero says. Rather than respond to a potential threat like a conventional alarm system—a so-called deterministic response, where almost any stimulus sets off a signal—their system would trigger a probabilistic response. They set the alarm to flash if it determined with a 70 percent probability that the moving object was an elk or a deer as opposed to, say, tumbleweed. False alarms plummeted, drivers began to trust the new system, and in the three months that they field-tested it during the winter of 2014, collisions dropped to zero.
Existing radar tracking systems could rarely distinguish between large birds and drones.
Around the time that Romero and Lamm were focusing on preventing accidents on the ground, more and more people started worrying about crashes in the sky. Once the province of military developers, then of rich folks who could afford the technology, drones soared into the mainstream in 2013 when Chinese drone maker DJI introduced the Phantom, the first consumer-priced unmanned aircraft system. It jump-started what Marke Gibson, the FAA’s drone expert and a former Air Force general, calls “the most fundamental change in aviation in our lifetime.” With hundreds of thousands of new aircraft navigating increasingly crowded airspace, Lamm and Romero noticed there were alarmingly few ways to keep track of the errant ones. What’s more, the radar tracking systems that did exist could rarely distinguish between large birds and drones, a problem similar to what they had encountered on the highway in Idaho. Seeing an opportunity to cash in on an emerging market, Romero and Lamm founded Black Sage in July of 2014 to adapt their wildlife-detection system to the new and more urgent problems posed by drones.
The FAA receives more than 100 reports per month of drones flying near aircraft.
The adaptation wasn’t as simple as taking their existing radar and camera equipment and pointing it skyward, though: Romero and Lamm had to write new software to process the ever-changing latitude, longitude, and altitude of an incoming target, all while taking into account the curvature of the Earth. Lamm wrote “slew-to-cue” algorithms so that whenever the radar picked up an incoming object, it would engage the camera, which then would track the object at a near-continuous 30 times per second. Later he and Romero added an infrared camera to detect the differential heat patterns between drones and the surrounding air. They headed to the scrubby hills above Boise to train the software, aiming the camera and radar at drones as well as the birds riding the thermals and the waterfowl in the wetlands below. For the drones and the birds, the system would measure acceleration, speed, heading, cross-section, surface area, whether the object had moving wings or propellers, and hundreds of other factors. “We didn’t have to know what makes these differences” between drones and other flying objects, Romero says. “The AI figured it out.”
By the summer of 2015 they had a system that could reliably detect an incoming drone about half a kilometer away, identify it, and stay locked on it regardless of evasive maneuvers. It was a breakthrough for them and a potential resource for anyone interested in keeping tabs on nearby drones. When the college security official invited Lamm and Romero to demo their system during the simulated nerve gas attack, he saw firsthand how the Black Sage system could track a drone. He also learned there was nothing that anyone could do to stop it.
You’d think shooting one down would be the easiest way to do it. After all, in 2015 a guy in Kentucky, pissed off that a drone was hovering over his property, grabbed his shotgun and shot the damn thing out of the sky. Simple enough. But it threw him into a thicket of legal trouble that he couldn’t escape for months. Under FAA rules, drones are considered aircraft: It’s just as illegal to shoot at one as it is to shoot at a Piper Cub, if for no other reason than you can’t control where (or on what or whom) a falling drone will land. The government has taken steps to prevent people from doing dumb things with their drones: Last summer the FAA released licensing and registration rules to compel drone buyers to learn how to fly responsibly. Drone manufacturers have taken actions too, integrating no-fly zones into the aircrafts’ GPS systems. Both measures are easy to get around, though, which explains why the FAA receives more than 100 reports per month of drones flying near aircraft—more than triple the rate it was seeing in 2014. No one knows what would happen if a drone got sucked into a jet engine, although computer simulations at Virginia Tech suggest that it would rip apart the engine’s fan blades in less than 0.005 second.
The Pentagon, spurred by reports that ISIS is using drones for surveillance and bomb delivery, has requested $20 million for antidrone research.
The problem goes well beyond aircraft. The Pentagon, spurred by reports that ISIS is using drones for surveillance and bomb delivery, has requested $20 million for antidrone research. Recently the Federal Bureau of Prisons posted a request for information on how to equip penitentiaries with antidrone systems (the better to stop drones from dropping contraband into prison yards). “Every prison, every airport, every facility with sensitive equipment outdoors, stadiums, amusement parks, racetracks … everybody is now worried about drones,” says James Williams, an aviation specialist at the international law firm Dentons. In short, what used to be a two-dimensional security problem—stopping intruders at ground level—has now become a three-dimensional one, as security breaches can come from above.
With US sales expected to triple over the next three years, drones are democratizing the air to an unprecedented degree, and Black Sage is only one of a handful of companies trying to solve the problem. One of the more promising, if flawed, systems in the works comes from British company OpenWorks Engineering, which has produced a bazooka-like device called SkyWall 100 that physically captures a drone with a net; the system won a recent competition for drone defense in urban areas, but it’s not effective much beyond 100 meters. In Holland, police have experimented with using eagles to attack drones, but they haven’t figured out how to protect the birds’ feet from the spinning blades, and the raptors have to be trained for months. In the fall of 2015, in their own first attempt to counter a drone, Lamm and Romero rigged a couple of ultra-high-powered spotlights to one of their tripods. When a drone approached, radar would detect it, cameras would track it, and with the touch of a button, 12 million candlepower of light would blind the drone and disable its video and espionage capabilities. It worked well at night, but when they demo’d the system for a customer in the Middle East, the desert sun rendered the lights useless against attacking drones.
Net Bazookas and Attack Eagles
Black Sage is one of a handful of companies trying to find solutions to the problem of errant drones. Here are some of the more successful—and problematic—technologies.
Guard From Above
Shortly after the high-wattage experiment, Romero went to an international security conference in Dubai in early 2016, where he met the owner of a company that makes radio jammers to protect armored vehicles in war zones. IEDs are often triggered by radio waves—via Wi-Fi or cell phone—and the company had produced a device that, mounted on a Humvee, broadcasted jamming signals at a broad range of frequencies in all directions. This got Romero and Lamm thinking about how frequency jamming could apply to their own efforts: Consumer drones are controlled through the public part of the radio spectrum (either 2.4 or 5.8 GHz). Blasting radio waves at those specific frequencies—jamming them—makes a drone deaf to its controller, which would cause the drone to return home or settle to the ground. A similar outcome would occur if you jammed the GPS frequency or what’s called the low-frequency L-band.
Frequency jamming is an elegant solution that doesn’t involve shotguns or trained animals, but it comes at a cost. Because these are public frequencies, jamming them disables other common electronic devices in the area, such as Wi-Fi, wireless home phones, and even garage door openers. Jamming GPS signals is even more dangerous—it can interfere with emergency responders and airplane-guidance systems. That is why jamming radio frequencies and GPS signals is illegal in the US. Still, Romero and Lamm thought that if they could jam only those frequency bands most commonly used in drone communication—and if they could limit their jamming to objects at which they have aimed their system—they could minimize the disruption to surrounding radio and GPS communications.
Since they couldn’t legally experiment near their headquarters in Boise, Romero flew to the Middle East to test out frequency jammers. After two and a half months of trial and error, Romero and Lamm created a new system that could bring down a drone with minimal impact on surrounding radio and GPS operations. Despite knowing that they couldn’t market it in their home country, Romero and Lamm pressed forward. “I know I’m going to regret saying this, but our thought process was, who cares about the States?” Lamm says. “We’ve got a $100 million customer in a hot, sandy place who doesn’t care about the FCC, and we have a solution they’ll love—so let’s do it.”
Lamm and Romero are understandably vague about where they test and sell their equipment overseas. There’s a spy-versus-spy element to the business, and you’re ahead of the game if your adversaries don’t know that you can counter their drone attack. A few times over several months, they called and updated me with their latest test results, and with each new dispatch they described various improvements and setbacks. Last summer I finally got a chance to see the Black Sage system for myself. On a remote hillside, I sat with Romero and Lamm inside a trailer set up as a command center. The drone-tracking gear consisted of two tripods: One held a cluster of eight Doppler radars resembling white iPads and, above them, the hi-def and infrared cameras; the other held the jammers—three white cylinders the size of paper towel tubes.
An assistant launched the quadcopter and flew it beyond eyesight, maybe a kilometer away. Moments after launch a white dot appeared on the radar-connected monitor. A readout confirmed that the object was a drone. Instantly the cameras locked onto it; and when Lamm zoomed in with the hi-def camera, we could see the quadcopter’s body and rotors. Lamm and Romero shot commands back and forth like a pilot and copilot. “Buzzer on,” Romero hollered. Lamm flipped a switch. A jammer emitted a storm of radio waves, blocking the control signal and paralyzing the aircraft. “Buzzer off!” Romero commanded, and the drone resumed the attack. “Buzzer on,” and it froze again. This time they kept the jammer engaged, and the drone settled to the ground.
Since then Lamm and Romero have updated their system yet again. A recent version, tested for an Asian counterterrorism unit last September, established a zoned system with a series of potential responses. If a drone approached within a certain distance of a prohibited zone, the system would jam its Wi-Fi and sever its connection to its controller. If the drone kept coming, that would mean it had been programmed to attack, and at that point the system would jam its GPS frequencies. “With zero human intervention, our system detected and identified the drone and took it down to the ground,” Romero says. “At that point, it was handshakes, smiles, and a happy customer.”
Though the Black Sage jammer includes a narrow-beam antenna to minimize frequency disruptions in the surrounding area, Romero and Lamm concede that using the latest version of their system in a crowded urban area could cause hundreds of businesses to lose their Wi-Fi for up to 30 seconds. It’s not something Lamm would use casually, even if the FCC allowed it. “It all depends on the threat level,” he says. “If you see a drone headed for an airport right now,” it’d be worth the risk of knocking out the surrounding Wi-Fi.
“I don’t think this is going to become real until we experience a catastrophe.”
It also depends on the environment. Lamm says he’d be comfortable using his system at an airport far from the city center or a stadium on the outskirts of town. Another good example, he says, is what Utah legislators had in mind last year when they passed a law that allows incident commanders at wildfires to use frequency jamming to neutralize any drones interfering with their work. The law is so new that it hasn’t been tested yet: Legal experts wonder what the FCC will do when an incident occurs, perhaps in the next fire season. (The FCC wouldn’t comment on Black Sage or the issue of frequency jamming.) Meanwhile, the FAA is hosting biweekly meetings with the FCC and other three-letter agencies to work out standards for what kind of antidrone systems can be developed and under what conditions they can be safely deployed. “The major issue is not just the technology, but the application of technology in a civil environment,” says Gibson, the FAA’s drone man. “We’ve never been in this position before; it’s the new frontier.”
Romero, Lamm, and others in their young industry hope that any new regulations will include a variance for emergency jamming. “I don’t think this is going to become real until we experience a catastrophe,” Romero says. Which would sound more cynical if he hadn’t witnessed a hundred kids pretending to die in a football stadium. Everyone then knew a drone was coming. The next time might be different.
Douglas Starr (@douglasstarr) is codirector of the graduate program in science and medical journalism at Boston University.
This article appears in the March issue. Subscribe now.