Hanns Tappeiner types a few lines of code into his laptop and hits “return.” A tiny robot sits beside the laptop, looking like one of those anthropomorphic automobiles that show up in Pixar’s Cars movies. Almost instantly, it wakes up, rolls down the table, and counts to four. This is Cozmo—an artificially intelligent toy robot unveiled late last month by San Francisco startup Anki—and Tappeiner, one of the company’s founders, is programming the little automaton to do new things.
<a href="https://www.wired.com/2016/07/cozmo-artificially-intelligent-toy-truck-thats-also-future-robotics/" class="clearfix no-hover"> <img width="600" height="338" src="https://www.wired.com/wp-content/uploads/2016/07/Cozmo-Lifestyle-002-600x338-e1468388562502.jpg" class="attachment-600-338-full size-600-338-full wp-post-image" alt="Cozmo-Lifestyle-002.jpg"> Related Video
The programs are simple—he also teaches Cozmo to stack blocks—but they’re supposed to be simple. Tappeiner is using Anki’s newly unveiled software development kit—an SDK, in coder parlance—that he says even the greenest of coders can use to tweak the behavior of the toy robot. And that’s a big deal, at least according to Anki. The company claims the SDK is the first of its kind: a kit that lets anyone program such an intelligent robot, a robot that recognizes faces and navigates new environments and even mimics emotions. With the kit, Tappeiner says, “we’re trying to advance the field of robotics.” He compares the move to Apple letting people build apps for the iPhone.
This is the kind of talk that accompanies just about every new contraption that emerges from Silicon Valley. But Anki has enjoyed an especially big dollop of hype. Big-name venture capitalist Marc Andreessen, who led Anki’s $50 million funding round in 2013, calls the company “the best robotics startup I have ever seen.” That may sound even stranger when you consider that Cozmo is a toy—a $180 gadget that might show up in a stocking at Christmas—but it also carries some truth. When it comes to intelligent robots, Cozmo represents the state of the art. Or thereabouts. The state of the art is ready for the world of toys, but not much else.
Anki
Tappeiner and his colleagues, a gaggle of PhDs who emerged from the robotics group at Carnegie Mellon University, will tell you much the same thing. Like so many others in the field, they admire the impressively mobile robots created by Google-owned Boston Dynamics, whose dog- and human-like droids radiate mechanical charisma. But Tappeiner questions how long it will be before these robots are genuinely useful. “Does it really make sense for us to create a farming robot—or will it take 20 years to really do that well? We can do this,” he says, nodding at Cozmo, “really incredibly well.”
What’s more, he believes, Cozmo can provide a seedbed for the future. Offering tools that even kids could use, a kit like Cozmo’s SDK could help breed a new generation of robotics researchers. But it also gives seasoned robotics researchers a path into the heart of this toy automaton, and that can help advance today’s work. “When we were in grad school,” says Anki CEO Boris Sofman, “you would have to pay $10,000 for a platform that had 10 to 15 percent of the capabilities of Cozmo.”
Nate Koenig, chief technology officer of the Open Source Robotics Foundation and a longtime robotics researcher, says Cozmo deserves some skepticism. “How expressive is it? Does it really respond to humans?” he says. “I would definitely be cautious before buying.” But he also says this kind of inexpensive yet malleable and at least marginally intelligent device can feed new avenues of research. “Any robot that you can program to have even some basic level of emotional contact with a person is a great research tool,” he says.
Don’t we already have robots that are far smarter than this toy? Not really. In the commercial world, robots often work on assembly lines or move stuff across warehouses. But these machines are pretty much hard-wired for specific tasks. Inside an Amazon distribution center, a Kiva robot just picks up a bin and moves it. It’s not teaching itself to play chess during its down time.
Yes, we’re moving toward robots than can respond to their environment and learn to do new things on their own. At a lab in Austin, Texas, IBM is plugging robots into its Watson AI services, which can understand and respond to questions and requests—at least in some cases. Last year, the US Defense Department’s Darpa research arm held an extravagant contest for intelligent robots. Google is now using technique called reinforcement learning—one of the techniques that helped bootstrap AlphaGo, the Google system that cracked the ancient gamer of Go—to teach robots how to pick up random objects. And researchers at the University of California, Berkeley, have used another key AlphaGo technology, deep neural networks, to teach machines how to screw one bottle caps. But all of these projects are really just experiments. The Darpa grand challenge was a failure, albeit a funny one.
Anki
In some ways, Cozmo is not quite at the forefront of robotics research. It doesn’t use deep neural networks, an AI technique that promises to reinvent robotics by allowing machines to learn tasks through the analysis of enormous amounts of data. It doesn’t have necessary on-board processing power for this, and since it doesn’t connect to the Internet, it can’t grab this power from distant servers. But using other techniques that aren’t as dependent on data analysis, Cozmo can recognize your face. It can pick up and move a set of blocks, even if they’re not carefully arranged. And by tracking certain events—Did it almost fall off the table? Did it just beat you at a game? Is it having trouble finding something?—it can mimic emotions. If Cozmo comes to the edge of a table, it might look scared. If it has just lost a game to you, it may pout and look to play with someone else.
As Koenig explains, this sort of “expressiveness” provides a foundation for others to build on. Thanks to the new SDK, researchers might even decide to connect Cozmo to other AI engines, including deep neural nets—a possibility Tappeiner says Anki itself may explore as well. Eventually, as hardware continues to improve, bots like Cozmo will be able to use AI like deep neural nets without needing to stay constantly connected to huge data centers in the cloud. Companies like Google and IBM are already pushing in this direction.
So, Cozmo is a toy. But it’s also the future. That future may be a long way off. But we have to start somewhere.
