On a fine day in a parking lot outside Boston, a man holding a smartphone steps in front of a 2015 Prius. The driver keeps pressing the accelerator. But a bunch of electronics hacked into the car brings it to a safe stop anyway—the system had already been tracking the pedestrian for some time using lasers and a camera. The vehicle won’t budge until the pedestrian is safely out of the way.
That (staged) near miss was a test of a prototype vehicle developed by researchers at MIT. They are trying to prove that there’s a different way to use robotic vehicles to improve people’s lives than the driverless taxi vision espoused by some automakers and tech giants, such as Alphabet and Uber.
MIT’s hacked Prius has the necessary sensors and software to drive itself without any human input. But it’s designed to test a concept dubbed “parallel autonomy,” where a human still drives, and the car’s computer only takes over when the meatbag behind the wheel is about to mess up. It’s the concept behind today’s safety features—like automatic emergency braking or lane departure warning—taken to its most extreme limit.
“Think of it more like a guardian angel than a chauffeur,” says Daniela Rus, the robotics professor leading the project. (It was partly funded by Toyota, which has said it is exploring the guardian angel concept itself.) Rus and her team think parallel autonomy could start saving lives sooner than cars to which we humans are just dumb cargo, because the technical and regulatory hurdles are lower.
Although many companies are testing prototype autonomous cars in cities, suburbs, and freeways, today’s prototypes still need constant supervision from humans in case they make a mistake. When regulators will be ready to let robots drive themselves—and us—on public roads is unclear. Nissan recently announced that because full autonomy was too hard, it would have robotic vehicles phone home for help from a human when they got out of their depth.
Rus applauds work on the holy grail of fully autonomous vehicles but says there’s no point just waiting around for all the wrinkles to be ironed out. More than 30,000 people die on U.S. roads every year. “Eventually everyone will get to autonomy, but the technology’s not ready yet,” she says. “We need to drive in rain, in snow, in heavily congested areas—this is an intermediate step we can take to make driving safer in the meantime.”
Asleep at the Wheel
The MIT team tricked out their Prius with a camera, 4 lidars, high-end GPS, and electronics that sense what a driver is doing with the controls. Like the prototypes tooling around Palo Alto (Alphabet’s Waymo), San Francisco (General Motors), and Pittsburgh (Uber), the car’s software uses its sensors to orient itself to a detailed 3-D map of its surroundings. It tracks moving objects such as pedestrians, cyclists, and other vehicles and plans what it would do next to keep driving safely. If what the human driver is doing with the controls doesn’t match with that, the computer takes control of the vehicle.
Rus admits that having the car take over is a weird experience—“Sometimes it feels a little bit sudden,” she says—but it works. As well as the tests with pedestrians, trials have shown it can slow down a driver going too fast or keep the car in lane if a person oversteers. So far the team are only testing their vehicle on private roads, at Fort Devens, an Army Reserve property east of Boston. (They have ready access to a more challenging proving ground: Boston ranks rock bottom in Allstate’s Best Drivers Report of 200 US cities.)
The guardian angel concept is literally millions of miles of testing behind programs fixated on the idea of making human drivers obsolete. But Wendy Ju, who studies the intersection of automation, humans, and design at Stanford University, says it’s reasonable to think it can gain ground.
One reason is that deploying autonomy this way avoids one of the biggest challenges facing companies trying to commercialize autonomous vehicles that do the driving. So far, even the most capable robot cars still need humans to take over sometimes, but we make extremely unreliable back stops. “People get sleepy in self-driving cars. It’s really hard to maintain vigilance,” says Ju.
That’s not to say that human frailties won’t cause problems for the guardian angel model. Ju notes that it will be important to figure out if some people act more recklessly when they know they have a robot driver looking over their shoulder.
Rus says that’s something her project will study, but that feeling they are in control of a vehicle should keep people responsible. She also argues that fewer people are itching to fully relinquish the steering wheel than all the talk of driverless cars might suggest. “I think we share with Toyota the vision that driving can be fun and people want to continue to drive,” she says.
Anyone that endured a congested commute today may find that hard to believe—and it’s hardly surprising a car company at risk of disruption might claim the status quo will endure. But results from Ju’s research giving people fake driverless car rides in a driving simulator offers some support. People asked to spend just 30 minutes in a staged robo-car report acute boredom. Given a choice, some prefer to take the wheel to pass the time. “Everyone thinks it’s going to be super exciting to be in a self-driving car but it’s actually going to be super-boring,” says Ju.