These days, AI-powered robots seem less a novelty than the logical confluence of a range of cutting edge technologies. Kentucky Fried Chicken’s new location in Shanghai features a humanoid clerk powered by Chinese search engine Baidu’s machine-learning smarts, for instance, and a Japanese ‘bot co-authored a short-form novel that nearly nabbed a national literary prize.
But there’s a realm — toys — that’s seen such AI revolution largely pass it by: While dolls like the Internet-connected Hello Barbie and WowWee’s CHiP robotic dog have come close to approximating the sort of emotional responses you would like to see from a sentient being, they haven’t quite nailed it to the extent that AI in other industries has. But Cozmo, a new robot from San Francisco-based company Anki, is different.
“We did it just how Pixar and DreamWorks interpret an animated character.”
Cozmo is best described as a palm-sized dump truck: It balances on two tank-like treads, and has a vertically articulated, cube-shaped head and tiny forklift. It’s got a built-in speaker, too, plus a myriad of sensors including a gyroscope, accelerometer, and cameras embedded within an LED screen above its motorized body.
It’s best observed in action. In anticipation of Cozmo’s launch, Anki’s releasing a series of videos it calls “Cozmoments” — sketches which demonstrate the little ‘bot’s ability to exhibit human-like behaviors. The first, “Best Friend,” shows the Cozmo playfully tapping the tail of the sleeping dog.
It’s the most advanced product Anki’s team has ever created, CEO Borris Sofman told USA Today. Development began in the fall of 2011, and the current prototype has 325 parts. That may sound like technological overkill for what amounts to a kids’ toy, but Cozmo leverages all those moving parts to convey emotions far beyond what your average robot is capable of. Leave Cozmo alone for awhile and it will become bored, with the pupil-less eyes on its heads-up LED display morphing to a game of Pong.
Put the robot in an unfamiliar place, by contrast, and it’ll explore exhaustively, stopping ahead of any precipices before taking careful, cautious glimpses over the edge. And all the while, it’ll play a shifting soundtrack that indicates its current emotion: chirping and cooing sounds when it’s contented, screeches and squawks when it becomes frustrated, and musical melodies when it makes a discovery or wants attention.
Cozmo’s games exemplify its human-like idiosyncrasies. Some aren’t quite so formal — leave the rules undefined and Cozmo will pick up and stack tiny objects on its own, reacting only when you poke it or knock it over.
Others, however, are a tad more traditional. Speed Tap, for example, has the robot attempting to tap blocks of the same color opposite a human opponent. When it loses, it throws a robotic tantrum, swinging its lift-like arm up and down and motoring haphazardly around. And when it wins, it celebrates with an audible chime and smug expression.
The key to crafting convincing mannerisms, according to Bosman, was imbuing a bit of movie magic. “We did it just how Pixar and DreamWorks interpret an animated character,” he told The Wall Street Journal. To that end, Anki brought on longtime Pixar animator Carlos Baena to fine-tune Cozmo’s facial expressions and movements. Syncing Cozmo’s four motors to Maya, a film animation software platform, and Anki’s custom-designed workflow allowed animators to see movements on Cozmo prototypes in real time.
Spontaneity, too, was key. “The reason we love the robots we see in the movies so much is because they don’t just respond with the same canned responses — they surprise you,” Anki President and co-founder Hanns Tappeiner told The Wall Street Journal. “Cozmo never gives you the same reaction twice. The head movement, the arm movement, the ways his eye animate on his face, or the inflection of his voice — the combinations change to create personality.”
“It’s taken years, but our technology is finally to a point where we can build a robot with personality.”
The final piece of the puzzle? A living memory. Using computer vision, a form of facial recognition, Cozmo catalogs human faces and reacts accordingly. The effect is eerily human-like: Cozmo’s eyes reflect tepid surprise when it spots someone new, and unbridled joy when a familiar face returns.
Perhaps as impressive as Cozmo’s depth of personality is the technology that drives it. The robot’s companion app for iOS and Android, which pairs via Bluetooth, packs the “emotion engine,” the code that Cozmo uses to handle decision-making and situational awareness. That same code dictates Cozmo’s animations and expressions, with other data processing taken care of in the cloud, via Wi-Fi. The app also acts as a centralized hub for managing Cozmo’s activity. It’s where you initiate games with Cozmo, and how you associate names with faces that Cozmo recognizes. It’s also how you wake the robotic companion from a charging-induced slumber, if need be.
Cozmo isn’t quite polished, yet — it’s shipping in October — but Anki is launching pre-sales of the AI-powered companion this week. It’s available for the discounted price of $160 (normally $180), and ships with a charging station and non-removable battery that lasts “an hour and a half” on a 10-minute charge.
Cozmo is an impressive bit of technology, but it’s undeniably barebones. The app only sports a handful of games right now, and Cozmo lacks any sort of conversation capabilities or advanced voice recognition. But those might be coming. Anki plans to develop “story-based” activities and “give Cozmo a few friends” in the coming weeks, Tappeiner told The Wall Street Journal. And given the company’s pedigree, that’s not too far a stretch: Its debut toy, the AI-powered Overdrive series of racing cars, garnered praise from such tech titans as Apple CEO Tim Cook and former Disney President Michael Ovitz.
“It’s taken years, but our technology is finally to a point where we can build a robot with personality — and bring some of the things you see in the movies to real life,” said Tappeiner.
Article originally published in July 2016.
Updated on 08-16-2016 by Kyle Wiggers: Added first advertisement for the Cozmo.
San Francisco startup Dimensional advertised for people to walk robot dogs through city streets. The New York Times reported that the job could include carrying a roughly 33-pound robot back to the office if it overheated. A lift home is an awkward requirement for a labor-saving machine.
Keeping the hardware running is only one challenge. Anthropic's robotics tests found that language models mostly failed when controlling joints directly, but did better when directing trained controllers. Giving an AI agent a body doesn't teach it to handle ordinary household objects.
Soon, a mattress protector may be more than a bed cover. Australian medtech company Sleeptite and researchers at RMIT University have received a US patent for FlexiSense, a smart bedding technology that tracks body position and posture to help caregivers identify when older residents may be at risk of falling.
The technology is being developed for Sleeptite’s in-bed monitoring system, REMi, and was recently on display at The MedTech Conference in Boston. A peer-reviewed study in Flexible and Printed Electronics says the sensors can also be made at scale without sacrificing performance or durability.
AI companies spend enormous amounts of time convincing us their technology is safe and secure. But what happens when those same companies become targets of increasingly sophisticated AI-powered attacks? Anthropic, the company behind Claude, is now facing that question. According to a Washington Post report, cybersecurity startup Cathedral discovered vulnerabilities that let its researchers access sensitive information in Anthropic's internal systems.
The researchers used AI agents to identify potential entry points in less than 12 hours. Interestingly, they didn't need to crack Claude itself. Instead, they exploited weaknesses in third-party business applications that Anthropic uses for everyday operations. Anthropic's security testing program disclosed the findings, which allows outside researchers to identify and report vulnerabilities.