Depending on who you talk to, self-driving cars are anywhere from a few years to a few decades away from disrupting our daily commutes. What most people would agree on, however, is that the technology necessary for fully autonomous vehicles is very nearly here—if it isn't already.
What's holding us back is, well, us.
While today's autonomous vehicles can to adapt to far more situations than even just a couple of years ago, it's dealing with humans that remains the biggest problem—having cars adapt to humans, and us adapt to them. That will eventually change when everyone is riding in fully-autonomous vehicles, but things are going to get considerably more complicated during the transition period in the interim.
"There are new issues that are going to crop up with automation," Dr. Anuj K. Pradhan, an assistant research scientist at the University of Michigan's Transportation Research Institute, told Motherboard. "We might be reducing the number of rear end crashes, but we may potentially increase the number of crashes where the car goes off the road because the driver doesn't react in time. It may create a new host of issues that we don't understand yet."
That's why he believes it's all the more important to study human behavior—what happens when a person is texting while a self-driving car needs him or her to retake control, for example, or what happens if that person falls asleep behind a self-driving wheel—in a simulator before we have autonomous vehicles on the road. To that end, Dr. Pradhan and the UMTRI team are now expanding the group's existing research to study exactly how humans behave in a self-driving car and how closely the cars may have to watch over us in order to ensure we're ready to take the wheel when the need arises.
In its 2013 guidelines, the US Department of Transportation defined five levels of vehicle automation. Levels 0 and 1 include most cars today, with either no automation or "function-specific automation" like electronic stability control. Level 2 involves "automation of at least two primary control functions," such as adaptive cruise control and automatic lane centering, while Levels 3 and 4 describe limited self-driving (like Google's car) and full automation.
According to Dr. Pradhan, "until you get to level 4, the driver is going to be a very important part of the entire driving loop."
While the University of Michigan has a 32-acre course called M City for testing driverless vehicles, Dr. Pradhan's research is being conducted in the university's driving simulator—a complete Nissan Versa in a lab with wraparound screens—both for practicality as much as necessity.
According to Dr. Pradhan, while the driving experience is virtual, the results they're able to get are close to reality. "There's some limitation to what kind of graphics can be displayed," he explained, "but in terms of driver behavior their reactions get very real very quickly. When we have people come sit in the simulator—obviously it's in a room, it's in a lab, but the first thing they do is put on their seatbelt. Their eye movements, their visual scanning patterns, are very highly correlated to how they scan in real life."
To track those movements, the UMTRI researchers have made some enhancements to the simulator. Those enhancements not only let them to do video monitoring of participants in the study, but perform physiological measurements of a driver's heart rate and EEG—and they believe that some of their methods will not only be useful for research purposes, but could be applied to future autonomous vehicles on the road.
"When you look at it from a practical point of view, you don't want to strap an EEG headset onto a driver who's on his daily commute," he added, "and so the most obvious monitoring technology then becomes non-invasive, remotely-mounted, video monitoring systems."
The video system used in the simulator is still able to monitor quite a bit on its own, including the driver's head position and eye movements—even eyelid closure, which can give an indication of fatigue or sleepiness. That information could then be used to alert the driver in an appropriate manner when the car needs to transition from autonomous mode to manual control.
Of course, monitoring research subjects in a lab is one thing. An actual self-driving car that's always watching its operator is another.
"Certainly there could be privacy issues with it, and that's a very real issue," Dr. Pradhan said. "I don't think that everybody will be very happy with cameras in their car, and that's something that regulators will probably have to deal with."
But for Dr. Pradhan, the primary motivation for making autonomous vehicles work is safety. "We have 30,000+ vehicle fatalities in this country every year," he said, "and automation is supposed to reduce that."
He's hardly alone on that front. The Canadian organization Mothers Against Drunk Driving has pointed to autonomous vehicles as a technology that could eventually help stop impaired driving. And a study from the consulting firm McKinsey & Company recently suggested that self-driving cars could one day reduce traffic accidents by as much as 90 percent.
While that may well happen eventually, the near-future of autonomous vehicles seems set to prove considerably more unpredictable. As with any new, potentially transformative technology, autonomous vehicles face their own set of challenges on the road to acceptance. But it's the breadth of challenges they face that sets them apart from most.
"The human factor is one of the biggest hurdles towards having an automated vehicle on the road," Dr. Pradhan said, adding that it is "one of maybe three or four." If he had to rank them all, he'd put human factor at number one, liability issues at number two, cybersecurity and privacy issues as number three, and ethical issues at number four—all of which, he noted, he ranks far ahead of technological issues. On that front, he said, we've "pretty much figured out what needs to be done."
As self-driving cars become more and more capable, those issues will also undoubtedly include more questions and concerns that many haven't yet considered. We may not only have to trust a self-driving car to watch the road and safely get us to our destination, but also trust it to watch us and judge whether we're fit to drive when it needs to hand over the wheel.