DRONE TRAFFIC CONTROL
The heavens are thick with drones…
At least 325,000 civilian UAVs were registered by U.S. citizens last year, with nearly 10,000 each day during the month of January alone. Now engineers and policymakers face an unprecedented question: How do you manage airspace for a growing swarm of autonomous vehicles?
Richard Kelley, chief engineer at the Nevada Advance Autonomous Systems Innovation Center (NAASIC), and his colleagues are working on an air traffic control and management system that can help keep American skies (and cityscapes) clear of potentially dangerous flying machines.
It’s the perfect job for the boundlessly energetic Kelley, a mathematician by training and engineer-cum-roboticist by passion.
BREAKGROUND: What’s the 1,000-foot view of the work you’re doing at NAASIC?
RICHARD KELLEY: NAASIC was created a couple of years ago as the result of Nevada’s Knowledge Fund, a $10 million state-run fund for educational institutions to conduct advanced research. Approximately $3 million went to the University of Nevada to establish what’s really the interface between researchers focusing on robotics, artificial intelligence, and machine learning and the business community.
As the chief engineer of that center, I’m responsible for doing the bulk of the in-the-trenches technical work, whether that’s consulting with companies doing R&D, or working with faculty members or students.
BG: Was there a particular moment when this issue presented itself? I keep thinking of the man who crashed his drone into the Empire State Building in February.
RK: There wasn’t a specific event involving a drone necessarily, but at the National Championship Air Races in 2014 — it’s mainly people flying really fast jets very close to the ground, quite literally air racing — we had partnered with Embry-Riddle Aeronautical University on the drone air racing section, which we called the “Drone Zone.” One of the groups present was a pair of researchers from NASA’s Ames Research Center in Mountain View who told us about this new project, Unmanned Aircraft System Traffic Management (UTM).
BG: Tell us about the system. I’m assuming this isn’t air traffic control in the way we normally think, with folks in an airport tower. How does it work?
RK: You don’t have to get too technical to appreciate the core of what’s involved here. Right now, NASA is building a cloud-based system for managing drones operating at low altitudes — think from the ground up to 400 or 1,000 feet. The basic concept is that when you want to fly a drone as part of a commercial operation, you connect to a system of UTM servers in the cloud and file a flight plan in the exact same way that commercial aviation does right now — describing where you want to fly, where you’re taking off from, where you want to be, how fast you’re going, and so on. The system takes that flight plan and checks it against a large number of constraints: Does this intersect with a national park? Are there other aircraft in the area that might collide with this flight plan? If it rejects the plan, you start over; if it accepts it, you’re good to go. That’s the current prototype.
“The thing I like to point out is the number of birds on Earth. The estimates range from 300 or 400 billion birds alive on the planet, but as we go about our business on a daily basis, we don’t notice these millions of animals.”
— Richard Kelley, Chief Engineer NAASIC
BG: How close are we to seeing the skies blotted out by autonomous drones?
RK: NASA has made it pretty clear that this is not ready for prime time, but the testing has been going very well. The testing is a lot of fairly standard software development testing for a cloud-based system — gauging the server, the client, and then communications. The one twist is that every couple of months we have flight testing, where we literally conduct operations in the air and verify that the system works with vehicles that are actually mobile. We’re going to see drones flying through New York City sooner rather than later.
BG: What does the future of our drone-filled skies actually look like to you?
RK: The thing I like to point out is the number of birds on Earth. The estimates range from 300 or 400 billion birds alive on the planet, but as we go about our business on a daily basis, we don’t notice these millions of animals . This is because, thanks to millions of years of evolution, birds have figured out how to fly safely and quietly and without really interfering with each other. The goal we all have in the UAV space is something similar: They aren’t constantly making noise, getting in the way, and so on. We may have a thousand times more aircraft in the airspace than we do today in the near future, but that doesn’t mean we’re constantly dodging drones falling out of the sky or ruining our lives. We’re trying to figure out how to make them as unobtrusive as possible.
BG: Cities obviously pose a unique set of challenges compared to rural and suburban areas.
RK: Cities are probably the most interesting case from a research perspective. One major question I think is interesting is that of unique microclimate-type weather — urban canyons, for example, affect temperature and air pressure, and those present challenges to automated navigation. It’s windy, in surprising ways! At this point, we haven’t done enough research to fly a large number of drones in urban canyons safely yet.
One more mundane challenge that’s interesting is that all drones right now, especially those used commercially, rely on GPS. If you are in an urban canyon, you don’t have visibility to a satellite, which means building in enough autonomy to allow a drone to operate in a city when it may not have a clear signal all the time, or the signal is so noisy that it’s off by several meters. Even if you input a flight plan, that’s based on GPS waypoints, and the drone will send back telemetry in-flight to the system to say “I’m here, I’m here, I’m here” so the system can validate its reported location as within the bounds of acceptability. If the drone all of a sudden is no longer capable of receiving a GPS signal, it essentially vanishes.
Designing a drone that can continue to navigate requires other sensor modalities: your accelerometer, your gyroscope, your drone-mounted cameras, a proximity sensor. Being able to navigate without GPS is going to be essential in this kind of world.
BG: You’ll also require a massive amount of computing power to handle an increasing number of drones in the air. How do these systems scale?
RK: The computing challenge is very nontrivial. If you look at most drones on the market, they have a single computer that’s a flight controller running a real-time system responsible for spinning the propellers. The newer systems will have a secondary computer that’s more like a computer in a desktop or mobile setting, responsible for higher-level intelligence, the number crunching and path-finding and whatnot. The trick is just figuring out a way to condense it all into a small-enough package that can actually fly.
Surprisingly, the FAA has stepped away from requiring specific equipment or having an air-readiness process for these small unmanned aircraft for commercial purposes. I could very easily see that changing as operations scale up.
One area where I definitely see standards changing in the near future is in communications. Right now, most drones in the U.S. operate over 2.4 GHz, the same spectrum as Wifi, which means that if you have a bunch of drones operating in an area with a ton of routers they’ll start to interfere with each other. As we see municipal WiFi become more common and the volume of drones in our airspace increase, it wouldn’t surprise me at all to see serious conversations around spectrum allocation issues. The FCC is currently working on appropriate ways to tackle that challenge, whether it’s a specific portion of the spectrum set aside for autonomous systems or alternatives to the WiFi area of the spectrum.
BG: You’re a roboticist at heart. Was there a particular moment that attracted you to working with UAVs in particular?
RK: This is the first time in history that we’ve seen widespread deployment of robots.
It’s funny: People don’t think of drones and robots as being very similar most of the time, but the reality is that they’re mathematically, dynamically and control-wise almost identical. If you think of drones as being a flavor of robot, the challenges we’re seeing — the boring stuff, like what frequencies do drones communicate on — are the first time we’ve grappled with these issues in the field of robotics. The problems we’re tackling right now with drones are the problems that in a decade we’ll have to solve with self-driving cars, or in two decades with humanoids.
We’re not just talking about spectrum allocation, but independent coordination, too. Right now we’re talking about air traffic control for drones — do we need a traffic control system for ground vehicles? Imagine a near future where humanoid robots are ubiquitous … they’ll probably share our sidewalks.