Drone racing is the future
Whirring drones will swarm around many homes this holiday season, leading to millions of miniature aviation accidents and frazzled nerves.
Imagine, though, if you could turn on the drone, and by itself, it could rise off the floor, map the area, and fly itself on any assigned mission? That’s the goal of the teams that competed for $1 million this month in the inaugural Artificial Intelligence Robotic Racing Circuit, called AlphaPilot for short.
Piloted drone racing is one of the most exciting of the emerging esports, with pilots wearing goggles that put them virtually onboard the aircraft as they fly through spectacular mazes at 70 mph. Pre-programmed drones can also compete on difficult courses, challenging a team’s navigational programming ability.
AI drones, though, must wake up on the launch pad, determine their location, identify the course’s gates and navigate to the finish line without human or GPS assistance. These teams don’t compete with quick reflexes or coding skills; they are designing the smartest brains possible for their drones.
“Everybody starts with the same drone, so everybody has the same platform on the same course. That means it’s an entirely AI and autonomy challenge,” said Chelsea Sabo, the technical lead on drone autonomy for Lockheed Martin, which sponsored AlphaPilot along with the professional Drone Racing League.
The arrow-shaped, 7-pound aircraft are less than 2 feet in diameter and equipped with four stereoscopic cameras. A small NVIDIA Jetson AGX Xavier processor powers the AI’s algorithms.
“We really want to see how autonomy is doing compared to humans because ultimately, we want our autonomy to be way better than how humans are performing,” Sabo told me at the final races in Austin.
The 50-yard course built inside an old warehouse was not nearly as complicated as those used to challenge human drone pilots. Large signs with QR codes marked the gates so the AI could identify them and follow the course.
Even then, the autonomous vehicles struggled to complete the flight. In the first heat of the championship, three drones crashed, and only one completed the course. One did not launch at all, while later, another drone crashed into three competitors on the launchpad.
While it may be easy to smirk, the crashes reveal the immense computational challenges. The drones are small and must carry all the processing power they need. They operate in three dimensions, which require sophisticated control systems, excellent machine vision, and rapid navigational calculations.
The teams can adjust their algorithms after every flight, but the challenge is akin to balancing a pin on the head of a needle. Tweaking one thing can throw off something else.
Racing began in October with nine teams from seven countries competing in Orlando. The competition traveled to Washington and Baltimore, becoming more complicated at every stage. The teams came from universities, companies and start-ups. The goal, as with most motorsports, was to test the limits of the existing equipment and software.
“Of course, we wanted to get to the top, but for us, it’s more about developing technology,” said Manan Gandhi, the captain of Icarus, a team affiliated with Georgia Tech. “We took a very high-risk, high-reward strategy with the hope of really getting ourselves familiar with the state of the art, so from that perspective, we actually gained a huge amount from this competition.”
A Dutch team, Mavlab, was the winner at the final championship in Austin. They represented the Micro Air Vehicle Laboratory at the Delft University of Technology in the Netherlands, and their drone completed the course in 12 seconds.
In a supplemental race, for a potential $250,000 bonus, Mavlab’s drone raced against a human pilot. But Gabriel “Gab707” Kocher completed the course in 6 seconds, proving that the human pilot is still superior to AI, at least for now. The AIs returning next year will be even better.
“Robotic sports, like drone racing, push the limits of speed and performance while creating thrilling opportunities to test and refine tech innovation for the real world,” said Nicholas Horbaczewski, founder and CEO of the Drone Racing League.
The real-world applications for those toys crashing around American homes include intelligence gathering, package delivery and search and rescue operations. Lockheed Martin, which makes drones for U.S. military and intelligence agencies, recognizes that autonomy is the next step for drones to reach their full potential.
So while drones whirring around the house may be annoying, they are the modern equivalent of the 1970s home computer kits that triggered the digital revolution. The kids who master drones today will own some of the biggest industries of the future.
Tomlinson writes commentary about business, economics and policy.