Kyle Vogt is not a good driver. He's more the type who steers with one hand, and appears to pay more attention to the conversation than to the road. One bright day last September, as he drove his Audi S4 near his San Francisco office, a Ford Mustang sped up and headed straight for his right rear fender. At the last possible moment, Vogt jerked the steering wheel and narrowly avoided a certain crash. "Close call," he said, laughing. Over in the passenger seat, I started breathing again.
Mere minutes earlier, though, Vogt was driving more safely. Or rather he was not-driving more safely, while demonstrating the handiwork of his company, Cruise Automation, which in early 2015 will become the first company to sell technology that enables cars to drive themselves. On a stretch of Highway 101, east of downtown San Francisco, Vogt had clicked a button between the front seats, turned a dial to adjust the speed, taken his hands off the wheel, moved his feet back from the gas pedal and brake--and then turned to look me straight in the face, while, at 60 miles per hour, the scenery ticked by.
In your first moments of riding in a self-driving car, your every instinct is to lunge for the steering wheel. But very soon you understand how the car does the thinking for you. Cruise's sensors, which monitor highway markings, constantly adjusted the steering to keep the Audi centered in its lane. When a truck swerved a bit close, the car instinctively slowed: Whoa. With Cruise's technology at the wheel, Vogt turned toward me more often--and his staffers in the back seat, head of operations Daniel Kan and engineer Rita Ciaravino, looked less worried when he did.
The drivers speeding past us likely had no clue what was going on. The only tip-off that this Audi was Cruise-equipped was the black bug-eye protrusion on the roof--a pod containing multiple sensors and cameras and connected to the computer in the trunk. A small apparatus that controls steering, acceleration, and braking was bolted unobtrusively beneath the steering wheel. The system can think faster than any human, "sees" without blinking, never gets tired or irritable or intoxicated--and is never, ever tempted by a smartphone. Cruise Automation will sell its RP-1 aftermarket kit, which will convert any Audi A4 or S4 to a self-driving car, for $10,000. Eventually, Vogt says, it will work with any vehicle.
Cruise has just 10 employees, and a lot rests on Vogt's broad shoulders. A redheaded 29-year-old from Kansas with tech-hipster scruff tracing his jawline, he was remarkably chilled out when I met him, despite the enormity of the task he faces and the fact he was getting married in a week. But then he's already been a successful entrepreneur two times over; most notably, he co-founded and wrote code for the company that became Twitch, which sold to Amazon in 2014 for just under $1 billion. When you hang out with him, you pick up a touch of Musk and Zuckerberg: the kind of founder who hangs back, cool and reserved, confident the world will come to him, and that time will prove him right. He's also spent much of his life working with the exact technologies required by a company like Cruise.
Still, to succeed Vogt must beat the business most synonymous with self-driving cars--a company called Google. (Perhaps you've heard of it.) Also, among others, Tesla, Ford, GM, and Audi. Those companies alone can boast a combined market cap of $700 billion, give or take a few hundred million. Meanwhile, in late 2014, Cruise Automation's technology was powering exactly two Audi S4s, one of which was Vogt's own. That doesn't make Cruise's task impossible. Self-driving cars, says Gartner Group vice president Thilo Koslowski, constitute a venture in which the likes of Cruise, should they innovate with "real ingenuity" in software and machine learning, "can offset the requirements of needing a big wallet." Still, he warns, "even the automakers don't have the resources to compete against Google."
None of this dents Vogt's ample self-assurance. "Now is the perfect time to take a shot at autonomous cars, and Google has just made it easier for us," says Vogt, using the technical term for self-driving vehicles. "In three years, you won't even bother buying a car unless it comes with Cruise."
Futuristic as self-driving cars may still seem, Vogt and many bigger players sense a huge opportunity. Their hunch is correct, says Koslowski, who in a recent report predicted that 25 percent of all cars will drive autonomously by 2030. More than 70 million vehicles are sold worldwide each year, he says, and "this technology will eventually find its way into all of them." And, he points out, there are approximately one billion cars in use worldwide today. A more conservative estimate comes from Jeremy Carlson, an analyst with IHS Automotive, who estimates 11.5 million self-driving cars will be sold by 2030.
While Google is not any entrepreneur's rival of choice, Vogt is right that it has made his task easier in one regard: The tech behemoth has proved the concept works, having famously tested a fleet of self-driving Toyota Prius cars and Lexus RX luxury crossovers while mapping every inch of northern California and racking up more than 700,000 miles of robotic driving. Such testing hasn't always gone smoothly. In 2011, a Google self-driving car slammed into another vehicle; Google later claimed a human was driving at the time. (Google's drivers, like Cruise's, keep their hands and feet at the ready while the car pilots itself.)
And Google keeps pushing onward. In May 2014, the company announced plans to test an extremely compact self-driving two-seater; a YouTube video shows a gray and white vehicle, a kind of ladybug on wheels, with an unusually facelike front grille. It won't have a steering wheel or brakes. Indeed, it's designed to go no faster than 25 mph, and appears intended for short runs on, say, college campuses, or in dense urban areas. (A Google spokesperson says the speed is capped at 25 mph for testing purposes.) What Google hasn't done, despite much publicity around its automotive efforts, is make any moves toward selling its self-driving car. "This is still a research project for them," insists Vogt. But Google has announced it will build 100 prototypes of that autonomous two-seater, working to refine its technology, and reports circulated in late 2014 that Google was seeking an automotive partner to bring that car to market, though it may take up to five years to do so.
Google denied my repeated requests to see a demo of that ladybug-like podcar. Its super-secret research lab, Google X, where the self-driving research takes place, is notoriously inaccessible to outsiders. Jason Short, a product design director at the drone startup 3D Robotics, has ridden in one of Google's Priuses, thanks to his boss, former Wired editor Chris Anderson. But he doesn't make that drive sound wildly appealing. "It drove like my grandmother on a Sunday morning," he says, unable to suppress a huge grin at the memory; a Google rep says such cars are intentionally designed for nonaggressive driving. Short's thesis: Google keeps prying eyes away from these efforts because those cars are far from being ready for prime time.
Vogt has spent much of his life thinking about robots. When he was 13, he built a 200-pound BattleBot for the bygone robot-combat competition--which became a Comedy Central show--and road-tripped with his dad to enter two BattleBot events. ("My bot was absolutely destroyed both times," Vogt says with a shrug.) Around that time, he built a miniature dune buggy. It used a webcam to read lane markings to autonomously navigate a predetermined route. He entered this device in his school's science fair, and won in a landslide.
Then, when Vogt was an undergrad studying electrical engineering and computer science at MIT, a friend found an abandoned safe in the basement of a campus building, and Vogt persuaded that friend to collaborate with him in making a one-armed bot programmed to turn the dial repeatedly, to crack its combination. "We let it run for 17 hours," Vogt says, until it opened the safe. While still at MIT, Vogt also programmed a Ford F-150 to drive itself across the Nevada desert, for the Defense Advanced Research Projects Agency (DARPA) Grand Challenge in 2005, though his team didn't make it past the qualifying round. Then Justin Kan found Vogt in 2006, when Kan was just starting online video site Justin.tv and seeking engineers. Only two engineers responded to Kan's come-ons. Vogt was one, and soon the two bonded over email--in part, Kan says, by discussing making an automated beverage server, though Vogt doesn't recall those exchanges happening until later.
In time, Justin.tv became Twitch and won traction as a place to watch live video gaming. (Cruise is funded entirely by Vogt and a small circle of investors, including Kan and other Twitch veterans.) While Vogt was still at Justin.tv, he helped develop the video-sharing platform SocialCam in 2011, which was sold to Autodesk in 2012 for $60 million. All along, busy as he was, Vogt kept building robots and dreaming of bigger things. In the summer of 2013, with Twitch already the kind of hit that would soon compel Amazon to purchase it for a massive sum, Vogt struck out on his own. And then landed on his really big idea--the one that drove us near San Francisco last September.
The idea of letting a car drive itself is now familiar enough that even state legislators are comfortable with it. In 2011, Nevada legalized self-driving cars. In 2013, Florida enacted a law that allows you to text while behind the wheel, so long as your car is cruising autonomously. Other state legislatures, from Michigan to Massachusetts, are considering similar measures. And now, other competitors are streaming in. In November 2014, one month after Audi track-tested a self-driving RS 7 at 150 mph, company chairman Rupert Stadler said Audi's automated cars would be on the road in 2016.
In addition to Audi, there are a few more potholes in view for Vogt. There's Google, of course, and also Tesla, which in late 2014 announced it will this year produce a new Model D that will include an autonomous-driving mode. Cadillac announced in September 2014 that its self-driving feature, known as Super Cruise, will be included in one of its 2017 models. Super Cruise will allow "hands-off lane following, braking, and speed control in certain highway driving conditions," says Dan Flores, spokesman for Cadillac's parent company, General Motors. "We're doing it because it's what customers around the world want."
Ford is working on a technology called Traffic Jam Assist, which will automate driving during certain stop-and-go situations, such as those encountered at rush hour on major highways. BMW has announced it wants to build a crash-proof car, and, at the 2015 International Consumer Electronics Show in Las Vegas, I rode in a self-driving BMW i3, sitting in the passenger seat of a driverless sedan while it piloted itself down a simple short course--maybe 100 feet long. The course was so simple, in fact, that the car barely had to think on its own, and it also never drove faster than 10 MPH.
It felt a bit like a ride at Disneyworld, or an airport shuttle. I was never worried, but I also wasn't particularly wowed. Also in Vegas at CES: the Mercedes F015, an unusual concept car--a protoype not yet in production--in that it does not have a driver's seat. Instead, you lounge in the back, where you can interact with the entertainment system using voice prompts and gestures, while you're driven around town; its heavily tinted windows create a cocooning effect. Mercedes drove it--or rather was driven in it--to the convention center. That said, there's no firm release date.
All these developments are being driven by continuing and exponential advances in computer technology. The smartphone you carry in your pocket is more powerful than the computer Vogt and his team at MIT used to make their F-150 self-drive 10 years ago. (A BMW rep summoned tha self-driving i3 in Vegas with a Samsung Gear smartwatch.) Advances elsewhere led to significant price drops for key components in automated car setups. The radar that Cruise uses to detect other cars on the road costs the company between $100 and $200--it's similar to the Doppler radar your local news channel uses to bounce a signal off cloud formations and detect weather patterns. Comparable older radars cost up to $70,000.
Such advances are why Mark Boyadjis, an analyst at IHS Automotive, says cars will eventually understand not just your immediate surroundings (a ball rolls in front of you) but how things happen in the real world (a child might be chasing that ball, so get ready). It's also why Cruise's initial product, the RP-1, carries with it a whiff of a beta test. While it costs $10,000, it only works as an aftermarket add-on to the Audi A4 or S4. At launch, it will function only on certain highways around San Francisco--the technology is easier to perfect in a geographically defined area--and at highway speeds. Vogt admits his company will have to expand quickly from the Audi models--within one year, he says--and work with other car brands.
Vogt is betting he can develop new products for cars and roads as you would for the Web--in other words, perfect a minimally viable product, and then gather data from users to improve the technology. (Cruise is set up to push software updates to its customers' cars.) Cruise's vision to expand offerings, by mapping greater swaths of the world and adding on other auto models, rests in part on the data its customers will gather. One factor in its favor is the growing and monied class of tech-savvy consumers who are willing to pay for novel, customized experiences--and for whom a standard car off the assembly line may pale next to the thrill of the next newest, shiniest thing. The other factor in Cruise's favor, besides getting to market first: its iterative approach. So far, Google's self-driving data has been based on a wholly closed system. The big car manufacturers are opting for far less ambitious plans.
A visitor to the Cruise Automation office--a converted garage off San Francisco's Gilbert Street--encounters a tight band of twenty- and thirtysomethings, earbuds in, working silently on computers, endlessly scrutinizing data from automated driving tests and fleshing out the algorithms that keep Cruise's cars in their lanes. (Like Vogt, several employees attended MIT.) When I stopped by, a six-foot whiteboard was on display on which someone had scribbled two lane markings and what resembled a quiz for advanced calculus--perhaps 30 equations, utter gibberish to the layperson.
Vogt, no longer an adolescent eager to show off his robots, blocked me from taking a photo of that whiteboard, and carefully avoided any explanation of automated driving algorithms. During a virtual demo on a computer, an engineer started explaining Cruise's sensor technology and its approach to lane monitoring--until Vogt quickly changed the subject.
This is understandable, because Cruise's cars interpret the world through its sensors, and the algorithms that manage the complex interplay of the inputs associated with motion--map routes, lane position, speed, cars, obstacles, road surface--are how it solves the math problems of driving. If Cruise Automation gets that math right--meaning, if the car interprets the sensor data correctly and drives you safely to work while you talk to your colleague about the weather or text your spouse about dinner plans--then the company might have a chance.
That alone may not beat a determined Google. But Google has a vast graveyard of failed products. The Nexus Q, a matte black orb that purported to stream movies to your television, never came out. The company bought Motorola to push smartphones out the door--before selling the division to Lenovo. Even some software projects go belly up. Google+ is still not exactly a household word. Google Wave, a radical attempt to redefine email and messaging--something only an engineer could love--was killed in 2012. Ventures aimed at setting up ad marketplaces for offline media, such as TV and radio, were quietly shuttered. "We look forward to working with many different partners to find ways to bring [self-driving] technology into the world safely," says a Google spokesperson. But it's hard to envision any world in which such vehicles would be available anytime soon.
Imagine the productivity unleashed if your car drove you on your daily commute. That's one huge attraction of the self-driving car. But the bigger societal problem all players are trying to solve, of course, centers not on texting pals while driving in rush hour, but on the roughly 35,000 people who are killed each year in accidents in the U.S. alone. Self-driving cars could represent the next leap forward in automotive safety. "People are injured or killed in car accidents that are preventable with the kind of technology we've developed at Cruise," says Vogt. "We'll look back at 2014 and realize how barbaric it is that we've let this go on for so long."
I thought about all this as I drove my rental car back to my hotel, after a day spent alongside Vogt--and, while pondering the possibilities, nearly rear-ended another car. It's too soon to know if Vogt is heroic or halfcocked. But a world piloted by cars under Cruise's control would have saved me from two heartstoppingly close calls in the span of a single day.
My rides in autonomous racecars: A visit with Stanford's self-driving pioneers
I clung to the shotgun seat while a self-driving Audi TT-S took corners far faster than most human drivers would dare, and Stanford professor Chris Gerdes, who runs the Stanford Dynamic Design Lab, sat at the wheel. Once strapped inside, we'd accelerated quickly--that TT-S could go 100 mph--and then Gerdes took his hands off the wheel. We raced into a turn and--good Lord--the tires screeched in protest, but Gerdes was trying to push its limits. "A robotic car is not always situationally aware," he shrugged, alarmingly casual--but the TT-S had reacted and quickly. Later, a grad student showed off another self-driving car's reflexes. We cruised toward a row of pylons at 40 mph--and then the steering abruptly lurched to the left and swerved right, as if it were trying to regain composure, before skidding to a stop. This was success--we didn't crash or flip. Still: Whew! Gerdes ran toward us, looking concerned, calling to the student about tweaking ... well, something or other. I'm sure I looked like I'd seen a ghost. But maybe I'd just seen the future.