Somewhere in northern Virginia, a man dressed as a car seat seeks the answers to vital questions about how autonomous vehicles interact with the public.
The fellow, who had nothing to say when confronted by a local NBC reporter on Tuesday, spends his days driving a silver Ford Transit Connect van around Arlington County. It requires a little skill to do this without moving one’s arms, but this goofy endeavor is done in the name of science, and builds on work done in recent years by similarly costumed researchers at Stanford University.
Car Seat Man is part of a Virginia Tech Transportation Institute Study into human-vehicle interactions—information automakers and tech companies like Google will find invaluable as they loose thousands of self-driving cars onto the country’s roads. The Institute confirmed that the guy inside that definitely-not-store-bought car seat costume and his shiny new van are part of its research effort. It didn’t have much to say beyond that, but the Institute notes on its website that it hopes to observe how humans react to robocars, and determine whether the folks making such vehicles should consider design tweaks to ease any tension or avoid any confusion.
This is a key question, because eliminating the driver eliminates many of the visual cues—eye contact, a friendly wave of the hand, an extended middle finger—that pedestrians, cyclists, and others often rely on as they navigate city streets.
“There’s a lot of debate right now about whether autonomous vehicles need signs of some sort to communicate their intent,” says Bryan Reimer, who studies autonomous vehicle and human interaction at the Massachusetts Institute of Technology. Figuring this out is tricky because everyone testing robocars on public streets puts an engineer behind the wheel to ensure the car doesn’t do something stupid. Those humans might make eye contact with pedestrians, or subtly nod their head toward another motorist. Their very presence could affect the way people respond to the vehicles in their midst. The only way around that is to, the degree possible, eliminate the person behind the wheel.
The Virginia Tech Transportation Institute declined to make anyone available for an interview, and a spokesperson wouldn’t say when it might wrap up the research. But the Institute plans to release the results. Arlington County officials know about the experiment, but aren’t involved. “We are interested in the research and glad to have testing done here, where we might learn more about the implementation of the technology,” says Deputy County Manager Jim Schwartz in a statement.
— Adam Tuss (@AdamTuss) August 7, 2017
Chicken Wire and Papier-Mâché
This isn’t the first time someone’s dressed up as a car seat for the benefit of society. Two years ago, a Stanford team led by Wendy Ju published a paper called “Ghost Driver” after spending nine hours (over the course of three days) on a similar study. “We were primarily looking at how pedestrians interact with self-driving vehicles, if people felt comfortable with crossing the crosswalk,” says Ju.
(By the way, the idea was inspired by a YouTube video in which a rascally dude dresses as a car seat to scare the milkshake out of drive-thru workers.)
Ju’s team made its first iteration of the car seat disguise using chicken wire and papier-mâché, then settled on “less pokey” materials. The head covering uses the same translucent material you see in Halloween costumes. “It helps to have someone on your team who can sew,” she says. Where the VTTI researchers are using an ordinary-looking van, the Stanford team put their car in costume, too, loading up a Volkswagen eGolf with “Stanford Autonomous Car” labels and the sensors an actual self-driving test car would usually have. But the costumed human did all the piloting.
“You have to get used to driving with your arms down,” Ju says, to keep them hidden from unsuspecting pedestrians. “You can’t have your arms sticking straight out. We practice that.”
While the Virginia experiment met with reportorial alarm, passers-by at Stanford reacted with a general “meh.” Sixty-five out of 67 pedestrians ultimately chose to stroll across the crosswalk in front of the car. When the researchers standing nearby chased them down to get their reactions, they discovered people for the most part trusted the vehicle. “They really feel so strongly that these are the rules of the road and [cars] have to follow them that they bet their life on it,” says Ju.
Now, people are probably much more trusting of autonomous vehicles in a techo-haven like Silicon Valley (where the vehicles are a regular presence) than Northern Virginia, especially if a silver van apparently without a driver isn’t labeled as autonomous at all.
But human reactions to cars in general are far from consistent. California and Vietnam have different driving cultures. So do California and Massachusetts, even Mountain View and Los Angeles. That makes the challenges of building human-robot communication systems even more fraught, and a topic of debate. Google has filed a patent for a messaging display for self-driving vehicles that might read “Safe to Cross.” Robotics company Drive.ai has experimented with roof-mounted displays with the kind of crossing signs you’ll see at any city corner. Swedish researchers built a self-driving display that smiles at pedestrians to indicate that it is safe to walk. Mercedes-Benz’s self-driving F 015 concept car conveyed the same message by projecting a crosswalk onto the ground. But it’s hard to say whether these solutions, or any solution, will work everywhere and for everyone.
In fact, researchers the world over continue trying to understand the most basic things about how humans operate with and around cars. “We need to understand how people within the driver seat communicate with pedestrians and cyclists to begin to think about how we augment automation to mimic some of those characteristics,” says MIT’s Reimer. Work like this isn’t just important for self-driving vehicle implementation, but for understanding how to prevent all kinds of crashes, no matter who’s turning the steering wheel.
The internet’s impulse is to ridicule, but the costume experiment has a point. “I think everyone should be glad that the manufacturers would want to know how the public is going to react to these cars before autonomous cars interact with people,” Ju says.
Some heroes wear capes. Others wear car seats.