Soon after I climbed into the back of a Google self-driving car recently, we drove through a yellow traffic light.
The car made it through the intersection with plenty of time to spare, while the light stayed yellow, and my fellow passengers almost didn't notice. A short time later, though, the car stopped for a yellow signal at a different intersection, and I found myself thinking I would have kept my foot on the gas.
Of course, I got a ticket a couple of years ago for blowing a similar decision.
That's the thing about the self-driving car: Google says its retrofitted Lexus RX 450h hybrid can make those calls with a degree of certainty that even a veteran driver may not be able to match. It's equipped with lasers and cameras affording a 360-degree view of everything around it; gyros and accelerometers to measure its own velocity; and processors that calculate the speed of every moving thing nearby.
Google plans to test about 200 of its two-seat self-driving cars this summer. The gumdrop-shaped prototypes (shown) will have manual controls for the test drivers to override the cars' autonomous driving systems, as required by California law. But Google plans to build most of the cars as fully autonomous — no steering wheel, no gas or brake pedal. California is expected to allow the operation of such vehicles on public roads — with a permit — by the end of the year.
— Los Angeles Times
The "self-driving" system also has the benefit of comparing what it sees with detailed digital maps, based on data compiled on previous drives around the city, that show everything from the exact positions of curbs to the location of lane markings and traffic signs.
During our 25-minute driving tour of some busy surface streets in Mountain View, Calif., Google's Nick Vanderpool explained that the car's onboard computer analyzed the variables and calculated that it had plenty of time to make it through the first yellow light.
The system also noticed that the second yellow signal came at an intersection where the cross streets were wider — which meant we'd have to travel farther to make it through to the other side, Vanderpool said. He was riding shotgun and monitoring the car's sensor readings on a laptop computer, while Google's Ryan Espinosa sat behind the wheel.
Espinosa wasn't steering, of course, except when we pulled out of the parking lot at the Computer History Museum, where Google showed off the cars to a group of reporters recently. Once the car's self-driving system took over, the wheel turned itself while he used both hands to adjust the air-conditioning vents — though a red override button was within easy reach.
(The demo followed Google's recent announcement that it's starting to spend more time with the cars on city streets, where the conditions are in many ways more challenging than on freeways.)
For much of the ride, I could have closed my eyes and probably wouldn't have noticed that Espinosa wasn't steering — except for a time when the car changed lanes abruptly, with a perceptible lurch.
"It's a lot smoother than it was a few months ago," said Vanderpool. He explained that the system moves the car in a manner that's most efficient, but like many new drivers, it has to be taught, through feedback from operators such as Espinosa and Vanderpool, that it can be more comfortable for passengers if the vehicle eases gradually into a turn.
"The car drives like a robot," he admitted. "The goal is to get the car to drive more like a human."
But not in every way. Google engineers stressed over and over that their first priority is to make the car safe — far safer than automobiles operated by humans. On the other hand, there are times when they've decided to rely on human judgment.
As we passed a police car that was parked on the roadside, I asked if the Google car is programmed to pull over when one of Mountain View's finest flashes the red and blue roof lights.
"We're working on that," said Vanderpool. But for now, the plan calls for the car's human operator to take over if that circumstance should arise.