Hard Drive: Self-Driving Cars Are Closer Than They Appear
By Lou Fancher
Before 2009, it was a secret. In 2010, it became an announcement. And from 2011 on, Google's self-driving car program has been an intriguing spectacle.
Now, five years on, Google cars regularly traverse Bay Area byways, sensing their surroundings and operating off internalized maps. At a late-January lecture at Livermore's Bankhead Theater, Google senior staff engineer Mike Montemerlo played a video compilation. A windshield-mounted camera showed faultless journeys: through dark mountain roads filled with big rigs and leaping deer; a residential, stroller-infested Mountain View neighborhood; a FasTrak toll booth; and highway construction sites. And what Bay Area driving test would be complete without a meander down pedestrian-rich Lombard Street? Throughout 10 challenging routes and 1,000 miles, the self-driving phenomenon performed like a robotic Galileo.
Originally composed of a tiny fleet of toaster-topped Priuses and one Audi TT, the program now boasts a dozen Lexus SUVs sporting Silicon Valley tech company Velodyne's Light Detection and Ranging system (LIDAR). Spinning at up to 900 rpm, the 64-laser rooftop whirligig creates a 360-degree point cloud — an enhanced "driver's" view. Other than the vehicles' high-tech hats, Google's autonomous cars hide their hardware: algorithm-loaded computers in the trunk, radar under the front hood, Vestigial Actuators where they always are (VAs are code for brake, accelerator, steering wheel). In the car's interior, the only aberrant features are a passenger with a data-collecting laptop and a Big Red Button (known as the BRB) — a clown-sized, electronics interruption knob to punch in a crisis. (Self-driving car language is rife with acronyms, applied to everything from roadkill to potholes to the car's sensitive circuitry.)
Beyond the geeky fascination, Google's program wields enough clout to earn commentary from transportation experts all over the Bay Area. At a sustainable transportation conference sponsored by Chevron in Concord in late January, Oakland-based Cambridge Systematics partner and ITS-Midwest Vice President Christopher Hedden said the interesting part isn't the technology, it's how we will live in the smart-car future. "Google's goal is to reduce the number of cars on the road. This will impact where you choose to live," he said. "Imagine an autonomous Winnebago. Get in after work at 7 p.m. and say, 'Take me to L.A.'" Turning to more serious matters, Hedden said connected vehicles and self-driving cars will greatly reduce drunken driver-related accidents.
Self-driving cars trace their origin to a Defense Advanced Research Project Agency invention, the DARPA Grand Challenge. Frustrated in its efforts to develop self-operating vehicles, the Department of Defense in 2004 dangled a $1 million prize (subsequently $2 million) for inventing a car capable of traveling 132 miles in the desert without a driver or remote control. The first year's winner managed to cover only 7.5 miles. But bright minds at places like Carnegie Mellon University and Stanford were turned on. In 2005, the second year, Montemerlo's Stanford team completed the course and won with "Stanley," a VW Touareg stuffed with computers. After 2007, DARPA's focus turned to robotics; the corporate world had taken the self-driving car challenge and run with it.Safety, Montemerlo says, is the No. 1 reason "you need this car in your driveway." With 32,778 auto-related fatalities in 2010 — 1.5 million worldwide, on average, per year — he said, "anything we can do to make driving safer can potentially save thousands of lives."
Smart cars save lives, he argues, by reducing human error. An autonomous vehicle doesn't get mad, drink and drive, fall asleep, text illegally, become legally blind or too old to drive responsibly — but remain too independent to stop — or practice playing trombone. Montemerlo showed actual photographic evidence of a driver practicing his instrument, earning a big laugh, but statistics from the American Automobile Association prove the sobering truth: Ninety-three percent of the 6 million annual crashes are attributable to human error.
Improved safety and economics also come from autonomous driving's efficiency. Total lane capacity on a freeway is estimated by experts to be 2,000 vehicles per hour. At peak capacity, only 15 percent of the space is used, according to Montemerlo. But that would change if sensors were allowed to command a car and decrease the "cushion" needed to drive safely. "If we could use more of the space, we could double the capacity of the road," Montemerlo said. Instead of adding lanes to handle congestion (costly construction), smart cars could operate more cars on existing roads, leaving more funds for road maintenance. Plus, greater efficiency would reduce the amount of time (and fuel) people burn up on the road: 30 billion hours per year, studies show.
Obstacles on the horizon do exist for Google's self-driving cars. They range from large issues, like insurance (if a smart car has an accident, who's at fault?) to legalities (which features are mandated for new cars? Is a solo-non-driver using a two-person HOV lane a violation?) to more straightforward concerns, like weather (rain is okay, but fog and snow are tough) and unusual circumstances ("Odd intersections, art bikes, Oscar Meyer Weinermobiles and other drivers doing weird things," Montemerlo explained). Google's cars rely on algorithm-driven maps that track the world like a video game, but creating models for flattened squirrels, abandoned mattresses, and kids on bikes is a challenge.
And then there are the competitors.
Richard Wallace, director of Transportation Systems Analysis at Ann Arbor, Michigan's Center for Automotive Research (CAR), says Google may not be any further ahead or fully invested than automakers. The difference, he writes in an e-mail, is that "automakers do not do R&D in the public (and their competitors') eye." Wallace said Audi, Mercedes, General Motors, Toyota,Nissan, and other manufacturers "are not far off," maybe as close as 2018. Despite heavy competition, Wallace says the role of Silicon Valley in the fast-moving field is enormous.
"Google has been extremely important in bringing automated vehicle technology into the spotlight and motivating the auto industry to accelerate its development," he says. The company's recent addition of radar, a step beyond its primary, 3-D imaging technology, will increase its self-driving car's viability, he adds.
Michigan, given its automotive history, is one of the "automated technology hotbeds" Wallace identified, especially in the development of connected, vehicle-to-vehicle technology. Silicon Valley, with its no-snow weather, venture capital momentum, and dedicated-lanes infrastructure, suggests a likely location for introducing mainstream drivers to the technology that will make them non-drivers. Montemerlo and Wallace predicted that lesser forms of automation (adaptive cruise control, lane-centering, blind-spot detection, parking- and traffic-jam assist) will continue to be incrementally released.
"The average vehicle on the road today is 11.4 years old," Wallace says. "Anything more advanced that comes out in maybe 2020 will be operating in mixed traffic for a long time to come. One day, you'll find you almost never control your new vehicle. I think that is about 2025."
But why stop at a car that can merely drive itself? Montemerlo says "self-aware" cars that can cruise "driverless" are likely 10 to 15 years out. Cars that can "reason about the world" are even further on the horizon, maybe even impossible. Until a car can recognize that the driver next to us is too preoccupied with his Slurpee, humans will remain behind the wheel.