If you had wandered up to the second floor of the Apple store in San Francisco on a Thursday evening a couple of weeks ago, you would have seen an unusual spectacle. The big wooden tables were cleared of the usual iPads and MacBooks, and in their place visitors had laid out large sheets of black vinyl printed to look like miniature speedways.
Zooming around on these high-tech tablecloths were squadrons of battery-powered model cars, each about twice the size of a Hot Wheels car. Groups of customers and Apple salespeople stood around the tables holding iPhones, cheering and hollering as the cars executed loop after loop.
The cars moved incredibly fast, hugging the curves with g-forces that would crush any human driver tiny enough to fit inside. Yet the cars weren’t running in tracks or slots, and weren’t hemmed in by walls; nor did any of them carry the computing power needed to drive on their own.
The real secret to the toy cars’ speed and agility was inside the players’ iPhones, which acted as the cars’ offboard brains. Five hundred times per second, each car sensed its position and speed, flashed the information to a paired iPhone over a Bluetooth wireless connection, and received instructions about how to adjust the precision motors on its back wheels to keep from careening off the track—or, sometimes, to deliberately slam into an opponent.
This is Anki Drive, and it’s likely to be one of the most sought-after holiday gifts this season, at least in the homes of families with multiple iPhones or iPads. The $200 starter kit and the $70 expansion cars are available only online or at Apple stores. (As Tolstoy said, all Apple families are alike; each Android family is unhappy in its own way.)
Anki employees are touring the continent this month, stopping at Apple stores in Boston, Chicago, Los Angeles, Dallas, and Montreal to show off Anki Drive. The startup famously debuted on stage this summer at Apple’s Worldwide Developer Conference in San Francisco, where it won an endorsement from Apple CEO Tim Cook as one of the first companies “using iOS devices and the iOS platform to bring artificial intelligence and robotics into our daily lives.”
Anki Drive is, indeed, something new on the scene. It’s partly a physical toy, partly an ever-evolving mobile video game, and partly an escapee from one of the country’s leading academic robotics institutes. And to hear co-founder and CEO Boris Sofman tell it, it’s also the first in what Anki hopes will be a series of surprising new applications for intensively software-driven robots in consumer life.
The common characteristics of these future products, in Sofman’s mind: “Relatively simple and elegant hardware; incredibly complicated software; and Web and wireless connectivity to be able to continually expand the experience over time.” My fellow tech journalist Austin Carr put it all perfectly in the headline for his recent Fast Company story: Anki Drive Is Racing Google, Not Other Toy Makers.
To learn more, I visited Sofman at Anki’s San Francisco headquarters recently. This isn’t some wacky hardware startup: befitting the $50 million in venture backing the company has raised from Andreessen Horowitz, Index Ventures, and Two Sigma Ventures, Anki’s 50 employees work in swank offices on the 15th floor of a highly secure downtown office tower. But there’s a comfortable game room set aside for demos and press interviews, and that’s where Sofman talked me through the genesis of the company at the Robotics Institute at Carnegie Mellon University, where he and his co-founders Mark Palatucci and Hanns Tappeiner were PhD students.
“It started as a way for us to moonlight and avoid our theses,” Sofman jokes.
Sofman’s own work at CMU focused on an autonomous vehicle: a $600,000 self-driving, off-road, all-terrain vehicle nicknamed Crusher, with applications in agriculture, exploration, and military reconnaissance. Palatucci worked on machine-learning algorithms for finding patterns in MRI data, while Tappeiner developed force-feedback haptic interfaces for mobile robots.
“The realization that we had was that in the Robotics Institute, there are amazing technologies all around, and people working on the most incredible things—pathfinding, perception, vision, manipulation,” Sofman says. “But almost all of these technologies were focused on pure research, defense, space, agriculture, or healthcare applications. Almost none of it was being focused on consumers. That was the impulse for us to focus on a consumer application, given how quickly the hardware landscape was changing. You can do things for a price point now that would never have been possible before.”
Specifically, Sofman says, it’s getting cheaper to solve fundamental robotics problems like positioning, reasoning, and execution. “You can take almost any problem in robotics and break it down into those three challenges,” he says. “It doesn’t matter if it’s an industrial robot moving a piece of steel or characters on a track in the living room. [Sofman calls the Anki Drive cars “characters”—more on that below.] You have to understand the physical world, use software to understand what you want to do, and then make it execute accurately.”
In the past, roboticists have solved the positioning problem using expensive equipment like motion-tracking cameras or lidar (the laser-based sensors on Google’s self-driving cars and many other prototype robotic vehicles). To provide accurate positioning in a $200 consumer product, Anki went in a different direction.
The black vinyl track in the Anki starter kit is imprinted with a microcode, written in a special ink that’s invisible to the naked eye. Each Anki Drive car includes an optical sensor—it’s actually the same one used in the front-facing camera of an iPhone—that samples these codes 500 times per second. The radio chip in the car sends the location information to the player’s smartphone, which then sends speed-adjustment instructions back to the two rear-wheel motors (they work independently, providing steering).
In other words, the cars only know where they are when they’re on the encoded racetrack; on an old-fashioned tabletop, they’re dead in the water. If you were an autonomy purist, you might call that cheating. But to Sofman, it’s an illustration of the practical compromises that will be needed to bring more robots into the real world.
“The art of thinking about the problem and how you can constrain it is the interesting thing,” he says. “We are constraining it not a huge amount, but making the problem much more tractable.”
What also makes a system like Anki Drive more tractable—and at the same time, more fun to use—is that almost all of the computation has been offloaded to
