Two decades ago, a semiautonomous, self-balancing, multipropeller aircraft would have been cutting-edge technology, hardly available at any price. Oh, how things have changed. On a recent walk through my favorite electronics retailer, I came across a tall stack of very capable little drones, selling for little more than it might cost for a nice dinner out.
Drones are one of several industries transformed by the manufacture of billions of smartphones. In the 14 years since Steve Jobs introduced the first iPhone, batteries, sensors, and electronics in general have grown both better and cheaper, thus finding wide use in a range of devices having nothing to do with mobile communication. Cameras and virtual- or augmented-reality gear also come to mind.
Drones stacked up like cordwood make for an impressive display at a shop, but as the numbers of them in the skies climb, opportunities multiply for bad actors or just naive operators to do all sorts of problematic things: invading people’s privacy, flying drones where they would threaten conventional aircraft, even weaponizing them.
The U.S. Federal Aviation Administration records more than 1.7 million registered drones—one reason why, at the end of 2020, it released a sweeping set of rules to manage their behavior. Within a few years, drones will need to be equipped with “Remote ID”: They will have to broadcast the equivalent of a license plate number, among other things. New rules include constraints on operations over people and at night. Commercial operators will no longer have to apply to the FAA for a waiver for such flights, but they will have to follow specific guidelines, based on the amount of risk their drones pose. Together, these regulations erect some needed guardrails in the skies.
These rules rely on compliance from the drone operator. But over the last few years, the linkage between a drone and its operator has grown less direct. At the higher end of the drone market, the operator no longer worries about keeping a drone in the air over a particular location: You simply program a flight path and let the drone execute those instructions in the way that it algorithmically determines.
Will a flight-planning algorithm know to avoid overflying a large crowd gathered, say, on a beach or at an outdoor concert? Would it even be able to detect people below it? Would the algorithm know to stop a drone when it reached the limit of the operator’s visual line of sight? Could it be fooled into operating at night, against the rules?
These potential problems arise from gaps in the drone’s ability to sense what’s around it. A drone knows everything about its position and speed and its battery levels, but nearly nothing about the larger world it operates within.
The same can be said for almost all autonomous cars. They can algorithmically stay in their lanes and avoid hitting pedestrians (or so we hope), but they can’t reliably read a temporary road sign specifying a reduced speed limit. These dangerous situations will become more common as the use of these autonomous systems expands. In a society soon to be crowded with billions of autonomous drones, package-delivery ground vehicles, and collaborative industrial robots, engineers need to focus just as much energy on building in safety as they do on expanding technical capacities. If we want to feel comfortable in a world with pervasive automation, we’ll need to put safety first.