In the basement of a Boston building, there’s a gymnasium where drones can play. Safety netting lines the walls, and the floor is covered with protective mats. Bright lights illuminate the room so a network of infrared cameras can capture the drones’ movements. One falls to the floor’s padded mats, bouncing along into a bumpy but safe landing. This is the dress rehearsal for the future of autonomous drones.
Over the past few years, many companies have proposed ambitious plans for unmanned aerial vehicles. Amazon, Google, Walmart and even Domino’s Pizza have all expressed interest in delivering purchases via carrier drone. Others imagine drones racing out alone to fight fires or manage crops.
But before we let drones loose, they need to learn how to drive on their own.
“Today, most things like this aren’t operating as autonomously as people tend to think they are,” says Kevin Leahy, at Boston University. “The more complicated, the more moving parts, the more vehicles you have, the harder it gets.”
The Boston lab aims to chip away at some of that difficulty. One recent experiment, led by Leahy, tasked a pair of UAVs with surveilling a particular square of the room. An algorithm controlled where they flew as they went about their work, tagging them out in turn so they could take a break at charging stations.
In other projects, squads of drones detect and fly around static obstacles in formation, or work together to build a picture of a patch of ground too large for any individual to monitor. Such programs could be valuable for missions that would require a team of vehicles to complete, such as larger-scale surveillance. The software would tell the drones where to go and what to do if there’s a problem, so operators could sit back and let them do the rest.
Leahy and his colleagues aren’t the only ones trying to figure out how drones can fend for themselves. In a video released last week by engineer Russ Tedrake’s lab at the Massachusetts Institute of Technology, a nano-quadcopter delicately weaves itself through an obstacle course of PVC pipes, plastic fencing and cotton string. Another, recently unveiled by researchers at the German Aerospace Center in Cologne, shows an autonomous aircraft flying at 75 kilometres per hour landing smoothly in a net on top of a moving car.
Right now, in the UK and the US, most drone pilots must keep their vehicles under their control and within their line of sight. But autonomy will be key for the success of projects like Amazon delivery, says Lawrence Brinker, executive director of the NUAIR alliance, a research consortium that runs drone test sites in New York, Massachusetts and Michigan. It’s crucial that drones not only know how to get where they’re going, but also how to navigate around obstacles on the ground and each other in the air.
“If you think about it, that’s about the only practical way to do it,” he says. “You’ve got to be able to tell your drone to go to 123 Smith Street and not hit the telephone pole out in front.”
At NUAIR, Brinker and his colleagues are tackling another aspect of the problem: what to do when someone’s drone unexpectedly wanders where it doesn’t belong, like near an airport. They have developed a sensor system to monitor incoming and outgoing traffic, alerting air traffic controllers to the presence of an uninvited guest. Possible contingency plans could include a method to quickly contact the drone’s owner, or a way of rerouting other vehicles out of the path of danger.
Others have suggested more extreme measures, like a tool that can disable the drone entirely, or even another UAV that captures errant drones in a net, like the one built by researchers at Michigan Tech University.
In one high-profile case in Kentucky last October, a man shot down his neighbour’s drone as it flew over his property. A local judge ruled in the shooter’s favour, although such a method may not be tenable for the future, says Brinker. “My guess is that, as that case works its way through the courts, we’ll find that’s not probably going to be an authorised method.”