Have questions, suggestions, or concerns?

Program Directors:

Mailing Address:

Siebel Scholars Foundation
1300 Seaport Blvd., Suite 400
Redwood City, CA 94063


(650) 299-5260

Let the Robot Drive - Featuring Chris Urmson '05

ARTICLE BY: Tom Vanderbilt

Wired Magazine

January 20th 2012

Let the Robot Drive: The Autonomous Car of the Future Is Here

The object, vaguely pink, sits on the shoulder of the freeway, slowly shimmering into view. Is it roadkill? A weird kind of sagebrush? No, wait, it’s … a puffy chunk of foam insulation! “The laser almost certainly got returns off of it,” says Chris Urmson, sitting behind the wheel of the Prius he is not driving. A note is made (FOD: foreign object or debris, lane 1) as we drive past, to help our computerized car understand the curious flotsam it has just seen.

It’s a Monday, midday, and we are heading north on California Highway 85 in a Google autonomous vehicle. In October 2010, when The New York Times reported that Google had built a fleet of self-driving cars that had already collectively traversed some 140,000 miles of California asphalt, it came as a shock, a terrestrial Sputnik. Now the cars, with their whirling rooftop laser arrays, are as familiar in the Bay Area as the company’s camera-crowned Street View vehicles. Indeed, the two are often confused, which is presumably why the words “self-driving car” have recently been plastered on this one’s driver-side door.

Anthony Levandowski, business lead on Google’s self-driving-car project, sits in the passenger seat, lanky and spectacled, wearing loud athletic shoes and clutching a MacBook Pro with a bumper sticker that reads “My other car drives itself.” Urmson, with the soft-spoken, intense mien of a roboticist who has debugged a Martian rover in the deserts of Chile, occupies the nominal “driver’s seat”—just one of the entities open to ontological inquiry this morning.
The Prius begins to seem like the Platonic ideal of a driver. It is better than human in just about every way.

The last time I was in a self-driving car—Stanford University’s “Junior,” at the 2008 World Congress on Intelligent Transportation Systems—the VW Passat went 25 miles per hour down two closed-off blocks. Its signal achievement seemed to be stopping for a stop sign at an otherwise unoccupied intersection. Now, just a few years later, we are driving close to 70 mph with no human involvement on a busy public highway—a stunning demonstration of just how quickly, and dramatically, the horizon of possibility is expanding. “This car can do 75 mph,” Urmson says. “It can track pedestrians and cyclists. It understands traffic lights. It can merge at highway speeds.” In short, after almost a hundred years in which driving has remained essentially unchanged, it has been completely transformed in just the past half decade.

Google isn’t the only company with driverless cars on the road. Indeed, just about every traditional automaker is developing its own self-driving model, peppering Silicon Valley with new R&D labs to work on the challenge. Last year, a BMW drove itself down the Autobahn, from Munich to Ingolstadt (“the home of Audi,” as BMW’s Dirk Rossberg told me at the company’s outpost in Mountain View, California). Audi sent an autonomous vehicle up Pikes Peak, while VW, in conjunction with Stanford, is building a successor to Junior. At the Tokyo Auto Show in November, Toyota unveiled its Prius AVOS (Automatic Vehicle Operation System), which can be summoned remotely. GM’s Alan Taub predicts that self-driving cars will be on the road by the decade’s end. Groups like the Society of Automotive Engineers have formed special committees to draft autonomous-vehicle standards. Even Neil Young is getting in on the act: Roboticist Paul Perrone has been busily revamping the rocker’s ‘59 Lincoln Continental to drive itself. “Everyone thinks this is coming,” says Clifford Nass, director of Stanford’s Revs Program.

As we drive the Google car—or are driven by it—I watch the action unfold on the computer monitor mounted on the passenger side of the dashboard. It shows how the car is interpreting the world: lanes, signs, cars, speeds, distances, vectors. The rendering is nothing special—a lot of blocky wireframe that puts me in mind of Atari’s classic Battlezone. (The display is just one of a host of geeky details—to change lanes, for instance, the driver presses buttons marked Shift and Left on a keyboard near the monitor.) Yet it is absolutely fascinating, almost illicitly thrilling, to watch as the car not only plots and calculates the myriad movements of neighboring vehicles in the moment but also predicts where they will be in the future, like high-speed, mobile chess. Onscreen, the car is constantly “acquiring” targets, surrounding them in red boxes, tracing raster lines to and fro, a freeway version of John Madden’s Telestrator. “We’re analyzing and predicting the world 20 times a second,” Levandowski says.

A car comes speeding along the adjacent on-ramp. Do we accelerate or slow? It’s a moment that puzzles many human drivers. Our vehicle chooses to decelerate, but it can rethink that decision as more data comes in—if, for instance, the merging car brakes suddenly. The computer flags a car one lane over, maybe 30 feet in front of us, and slows imperceptibly. “We’re being held back by this guy because we don’t want to be in his blind spot,” Levandowski says. A bus suddenly looms next to us. “Even if you can drive in the center of the lane, down to the centimeter, that doesn’t mean it’s the safest route,” he says. And so the car drifts just a bit to the left to distance itself from the bus. “If you look at it, we’re not actually driving center, though we’re still not driving as bad as he is,” he says, pointing to a gray SUV ahead that’s straddling two lanes.