Google's self-driving cars are by now famous even though the Internet company is not selling any of the futuristic, koala-faced vehicles just yet. Experts say the robotic cars are not quite ready to take passengers down the millions of miles of American roads.
The MIT Technology Review approached a number of researchers in autonomous driving to ask about what they thought of the self-driving cars being developed in Google's Mountain View home in California. The experts all agree on one thing: Those self-driving cars still have a lot to learn. In fact, they need to learn so much they say it is still much better to keep the steering wheel and allow a human driver to take over until Google's engineers come up with new ways to address the cars' major limitations.
"This is a very early-stage technology, which makes asking these kinds of questions all the more justified," says Michael Wagner, a robotics researcher at Carnegie Mellon.
Chief among these questions is how Google plans to collect information that will be programmed into its cars. Technology Review writes that each time a Google self-driving car gets ready for a trip; engineers make a handful of preparations, including extensively mapping the car's exact route. It is not unlike taking a Street View map of a road for Google Maps, but with the self-driving car, the mapping is much more detailed.
"Data from multiple passes by a special sensor vehicle must later be pored over, meter by meter, by both computers and humans," writes Technology Review. "It's vastly more effort than what's needed for Google Maps."
Another common obstacle the self-driving cars still have not learned to overcome is the weather. While human drivers have something called a brain sitting between their ears to help them navigate their vehicles through rain, fog or snow, Google's engineers still have to figure out how to teach their driverless vehicles to do things such as go slow in wet pavement or estimate the lanes when the road is covered in snow. In fact, Chris Urmson, the head of Google's driverless car team himself, admits the cars have yet to be tested in wet conditions because of concerns posed on the safety of the testing engineers.
Moreover, MIT notes that the cars are not 100 percent equipped to handle mapping omissions, such as road constructions, temporary stop signs or lane diversions. Although the vehicles are designed to recognize physical obstacles, they do not know how to respond to certain obstacles. For example, if new traffic lights were put up, the car would slow down as it approaches the stoplight but will not know how to obey it. The cars also would not be able to spot things such as potholes or an uncovered manhole.
The most unnerving observation, however, is that Google's driverless cars still have to recognize pedestrians as humans, not the "moving, column-shaped blurs of pixels" that MIT says the cars see them as.
"The car wouldn't be able to spot a police officer at the side of the road frantically waving for traffic to stop," says MIT.
Hopefully, however, Google finds a way to get around these obstacles before they hit the roads in time for Urmson's 11-year-old son's receiving his driver's license five years from now.