A question kept ringing through my mind as I took my first ride Monday in one of Google’s prototype self-driving cars — Is this newfangled robo-car a better driver than I am?
I thought about taking my driver’s test many years ago at the Los Gatos DMV, my hands nervously gripping the steering wheel of my mom’s cumbersome Ford Aerostar while a heavyset bureaucrat checked off my mistakes. I didn’t scan my mirrors frequently enough; one of my turns went a little wide; I hesitated to remember my turn signals.
Nevertheless, I passed and became a newly minted American motorist — one who still remains alive to drive another day despite my share of errors.
Many aren’t so fortunate: Each year approximately 30,000 people are killed due to roadway crashes, the equivalent of a 737 airplane falling from of the sky every four days. These grisly numbers are fresh in my head because they come up eventually in most conversations with a Google spokesperson while talking about the autonomous-vehicle program.
They bring up the crash statistics to make a fair argument. If human mistakes are to blame for all those fatalities, then isn’t it worth trying to design a system that will never lapse in judgment? Wouldn’t the nation’s roads be safer if all cars were scripted to abide flawlessly by the DMV handbook?
That autonomous car revolution still remains far off, but it is inching closer by the day, and its crucible is the streets of Mountain View. The Google team, which has already logged roughly 1 million miles on local roads, announced it would begin ramping up its testing this week with a new line of 25 custom-built prototypes navigating around the city.
To mark the occasion, the Google team invited the Voice to take a spin and see its automated cars in action. We met Monday at the Google X campus off Mayfield Avenue and proceeded to pile into one of the older Lexus prototypes, which the company has been testing along local streets for about a year and a half. We weren’t getting a ride in one of Google’s new custom-built car because, with only two seats, those models were too small to fit the five of us: me, my photographer and three Google team members. The differences were minimal since both models featured essentially the same software and sensor array, we were told.
With everyone seated, the car’s feminine voice intoned “auto-driving”, and the ghost of an unseen algorithm began turning the steering wheel on its own, maneuvering us out of the parking lot with unnatural grace and the cautiousness of a brake-happy grandma.
Sitting in the front seat, program manager Shyan Izadian showed a visual display of the car’s computer brain running on his laptop. Outfitted with more than a hundred sensors, including lidar, lasers and cameras, the car’s computer had a panopticon-like vantage of everything in its immediate vicinity. The program visualized everything in crude vector graphics resembling a 1980s arcade game, showing a cyclist as a blocky red polygon, and a pedestrian as a yellow box.
If anything, it seemed safe almost to a fault. I wasn’t sure if the car ever reached 25 mph, the top speed for this phase of street testing. At any point of confusion, the car by default came to a soft halt. A parked car that jutted out a little into the street? Brake. A long tree branch drooping into the street? Brake. Like something straight out of a driver’s-ed class, a child’s ball even bounced in front of our car as we were passing Thaddeus Park. Sure enough, the car braked.
Our short joyride left me with little doubt that the Google car would pass, or maybe even ace, a standard DMV driving test. That’s not to say I’d want to drive behind one if I was in a hurry. Perhaps to comfort my own ego, I began racking my memory to think of a time when being a leadfoot driver was safer, if not just more fun.
Obviously, the Google design team is emphasizing safety as a top priority at this phase, explained spokeswoman Maggie Shiels. There was simply no way to anticipate everything that could happen on the road. She cited one famous incident when one of their self-driving cars encountered a woman in a wheelchair armed with a broom was chasing a duck down the middle of the street. Even a human driver is frequently baffled by these kinds of situations, she said.
“You could sit in a conference room and never come up with these kind of scenarios,” Shiels explained. “The important thing is this is a car that never falls asleep, loses its attention or has a hangover.”
Coming to our first traffic signal at Rengstorff Avenue, the car obediently waited at the red light like other drivers. As the light turned green, the car waited a few seconds then cautiously crept out into the intersection for a left turn. The car zipped us back to the Google parking lot, taking its final curve a little too sharply and bumping its rear tire on the curb.
Admittedly, some kinks are still being worked out. Google program manager Greg Hanabusa showed how he could take manual control of the vehicle at any time by fiddling with the pedals or steering wheel, in accordance with current state DMV rules.
Walking into the Google campus, we were led over to one of the new two-seater prototype cars, which bears a striking resemblance to a koala bear, and will hit the road this week. Systems engineer Jaime Waydo, who previously worked on the NASA Mars Rover project, described how the new model was designed from the ground up, jettisoning accessories unnecessary for a car that steers itself.
Sitting in the driver’s seat, it felt at first like burglars had stripped the car of its parts. There was no steering wheel, no pedals or mirrors and all the car’s controls were all located on the center console between the two seats. Driving would only require pressing the console’s prominent green button, and entering a designation into your smartphone, or perhaps in future versions, just saying it out loud, she explained.
Pressing the same button while en route instructs the car to immediately find parking. The other controls were fairly standard (windows, locks), except for a tech support button that immediately phones someone at Google for help.
The final event in our whistle-stop tour was a chance to speak with Chris Urmson, the lead visionary of Google’s self-driving car effort. For more than a decade, Urmson has been tinkering with the ideas and challenges of autonomous vehicles. As a professor at Carnegie Mellon University, he participated in the U.S. Department of Defense’s 2004 Grand Challenge, which set the goal to design a self-driving vehicle that could travel 150 miles. None of the robot vehicles completed the journey, although Urmson’s team made it the farthest — 11.7 miles — at which point their vehicle drove through a couple fence posts and got stuck on a rock.
“The media kind of crucified everyone involved: ‘You only went 11 out of 150 miles!” he said. “There’s always been this challenge of getting this complex system to work, and the hardest thing is always to understand the world around the vehicle.”
But after that humble beginning, the technology evolved by leaps and bounds, Urmson said. At the next Grand Challenge in 2005, five teams including Urmson’s were able to complete the course. Later contests challenged participants to design autonomous vehicles that could maneuver around moving obstacles, obey traffic signs and stay on the right side of the road.
It is getting very close to the point where a consumer model would be ready for production, Urmson said, although he couldn’t specify how much longer that would take. He is setting a personal goal to have a self-driving car available for consumers before his 11-year-old son reaches driving age.
Recently in online discussions, many people have wondered what a self-driving car would do in an dilemma like the classic trolly-car problem. In that scenario, a trolly car is barreling down the road and going to kill five people, and the operator must choose whether to divert the car and saving those five people, at the cost of killing another bystander.
Urmson gave assurances that scripting how a self-driving car would function wasn’t much different than the snap judgments drivers routinely are forced to make on the road. The best precaution his team could make was to avoid no-win situations, he said. Outside of that, it was matter of minimizing the crash outcome, he explained. The self-driving computer would put a priority on avoiding hitting cyclists and pedestrians, whereas a slightly lower priority would be given to dodging other cars or static objects.
“People have been arguing about the (trolley) problem. Since there’s no right answer, we have to frame it in a way so we have a viable solution,” he said. “If it’s on the road, it’ll be incredibly safe, but it will be fallible and other people on the road will make mistakes as well.”
Citing the high number of people killed on the roads, Urmson expressed complete confidence that a robotic system would ultimately be a vast safety improvement over human drivers. For now, his team is intently studying how human drivers, cyclists and pedestrians react to their new cars. Stickers on the back of all the new Google cars encourage passersby to give comments on a new website. The early response from Mountain View residents at this phase will play an integral role in how the cars are further developed.
I asked Urmson what message he’d like to get out to the locals. Mountain View is home for Google, and being part of the local community was extremely important, he said.
“There’s huge potential for this technology, but if it doesn’t land right … it won’t get adopted and we won’t have these society benefits,” Urmson said. “The whole spirit of what we’ve tried to develop here is something that should fit into the wider community.”
Email Mark Noack at mnoack@mv-voice.com
