It’s Alive! My piece on Pleo in Wired magazine

I’m coming late to this one, but last month Wired magazine published a piece I wrote about Pleo — the new toy that’s about to be released by Caleb Chung, the guy who invented the Furby. As you might recall, last week I blogged about the interesting moral implications of the 3D avatar-based recreation of the Stanley Milgram “shock” experiments; since the experiment illustrated that people do indeed emotionally bond with virtual lifeforms even when they know the lifeforms aren’t real, some of the moral issues — what does it mean to mistreat, or to love, a robot? — are going to be raised by Pleo, the most lifelike commercial robot yet.

The story is online at Wired’s site, and a copy is archived below!

It’s Alive!

Say hello to Pleo. From the guy who brought you Furby, it’s a snuffling, stretching, oddly convincing robotic dinosaur. You are so going to want one.

by Clive Thompson

When I first meet Pleo, the tiny dinosaur is curled up on a kitchen table, its long tail and big head pulled inward. It’s snoring quietly, emitting a strangely soothing sound, almost like the amplified purring of a guinea pig. I’m tempted to reach out and touch it — but it looks so peaceful, I can’t bring myself to disturb it.

Then I realize what I’m doing: I’m worrying about waking up a robot.

Caleb Chung seems to understand my reluctance. “It’s OK,” the toy’s inventor says, motioning to the little green lizard. “You can touch him.” But before I do, Pleo wakes up on its own, fluttering open its doelike eyes and lifting its head. There’s a barely perceptible whizzing as its 14 internal motors spring into action and it struggles upright, stretching itself to get the kinks out. “You know, all your dogs do that,” Chung says as Pleo begins to poke around the table. “They wake up in the morning and go ‘ummmm’ — just like that.” The dino lets out a long, creaky honk.

“I think he wants to play,” Chung suggests, so I tentatively stroke the nubbly rubber skin on its back. It moos happily. A laptop on the kitchen table is monitoring Pleo’s internal state. As I trigger the touch sensors embedded in the toy, its “arousal” numbers start rising: 16, 23, 27, 28. It’s like a Matrix view of Pleo’s subconscious. I poke its left leg, and it cranes its neck curiously to see what just happened. I’m impressed. This feels less like interacting with a piece of machinery and more like playing with a kitten.

Chung knows how to create emotional connections to toys. Ten years ago, the bushy-haired, hyperkinetic inventor conceived Furby, selling more than 40 million of the yammering gremlins in a worldwide craze that launched the now-booming industry of robotic pets. A string of artificial companions have since trundled off the production line: the FurReal cat, the Roboraptor, the Robosapien, the Aibo and its litter of me-too electronic pooches. Household robots have arrived — not as servants doing our laundry but as helpless, babylike things that demand we take care of them.

Still, all of them have acted like, well, robots. But Chung, now 50, has a different idea: He wants to create “an artificial life-form” — something that looks eerily alive and is affected by its environment. Pleo begins as a baby, and its personality is forged by how you treat it. If it uses a high-pitched squeak and you feed it, it will learn to repeat that noise to get fed. Be nice to it and it will become mellow and friendly; mistreat it and you will evolve a bitter, annoyed robot. Theoretically, no two Pleos — Pleii? — will end up with the same personality.

The early buzz has been deafening. When Chung and his company, Ugobe (pronounced you-GO-be), showed off a blinking, newborn Pleo to the DEMO new-technology conference in February 2006, more than 500 attendees sang “Happy Birthday” to it. Ugobe will begin taking preorders for Pleo on its Web site on February 3. What gives Pleo its emotional hooks and makes it seem so much like a sentient pet is how it moves. Chung has managed to faithfully capture graceful, animal-like locomotion. There is none of the jerky machinelike quality that mars most bots. “Nobody has ever done that,” Chung says. “They’ve spent $2 million and a year trying to get their robot dog to walk, and it’s still like this,” he adds, suddenly twisting his body into a scarily accurate imitation of a stiff-limbed automaton.

“It’s like, ‘Look at me! I’m a robot! I’ve got gears and motors inside me! Zzzzt! Zzzzzt!’” Then, for good measure, Chung breaks into a disco “robot” dance, and the programmers clustered around Pleo start chuckling. That’s when I realize I’m looking at Ugobe’s secret weapon: Chung’s uncanny physicality.

Because how do you create the first robot that seems like it’s truly alive? By starting with an inventor who knows how to move.

CHUNG BEGAN HIS CAREER as a mime. In his early twenties, teaming up with comedian Gary Schwartz, he performed everywhere from cruise ships to The Alan Thicke Show. A short, tightly muscled guy, Chung was famous for pulling off Cirque du Soleil-type feats. In one act, he pretended to be an astronaut lifting off into orbit: While seated on a chair, he slowly raised himself up by his hands, inverted his body, and rose into a handstand.

“He looked like he was floating in space,” Schwartz says. “He blew people’s minds.” Enamored of special effects and handy with tools, Chung made a sword out of spare sofa parts and duct tape — then used it as a calling card to break into stunt work. Being short and trained as a mime got him inside high tech movie costumes, including an orangutan suit for Greystoke: The Legend of Tarzan, Lord of the Apes. To the studios, he was a double threat: He could perform like a monkey and then fix the robots when they broke.

“What I learned from all the mime work and the suit work is that motion creates emotion,” Chung says. “How you stand, how you move, is a big communicator. We take it for granted, but it’s crucial to what makes us seem ‘normal’ to each other, right?”

In the mid ’80s, he left Hollywood to work in the R&D division of Mattel — “toy college for me,” he says. It was not a felicitous match. Mattel wanted him to crank out action figures from popular movies; Chung wanted to produce art. He posted a sign over his cubicle that proclaimed THESE THINGS ARE NOT TOYS and began dreaming of making a robot so realistic that people would treat it like a household pet. His earliest sketches were of dinosaurs.

“They have this long neck and tail; they’re very expressive,” he explains. “Plus, all the people that don’t even like toys are going to say, ‘Cool dinosaur.’” He made a rickety prototype by repurposing a toy originally designed as a He-Man accessory. Mattel executives were intrigued — but recoiled when they discovered it would need eight motors; those cost $1 apiece, and a $30 toy couldn’t include more than one or two. He told them they should build it and charge more. They told him he was crazy and killed the project.

Disillusioned, Chung later left Mattel and went freelance, devising and selling inventions, like an “action man” and an automatic hair-curler. But he still hankered to develop a virtual pet, and in 1997 he brainstormed an idea with David Hampton, a programmer friend. They called it Furball: a tiny, tribblelike thing that would have eyes, ears, and a mouth — just enough to create the illusion of sentience (“the simplest haiku of a life-form you could get,” as Chung puts it). To keep it from being too expensive, Chung worked out a cunning set of gears that would drive the entire toy using a single motor. Tiger Electronics loved the concept and commissioned it.

Everyone knows how the story ended: Furby came out in 1998, and holiday consumers went berserk, buying $1.2 billion worth of the $30 toy. Chung made more than $10 million in royalties. He now had the freedom and money to do precisely what he wanted.

And that dinosaur was still on the list.

WHEN I FIRST MEET CHUNG and his eight developers in Boise, Idaho, they’re in a mild state of sleep-deprived panic. Hoping to snare $8 million to finish their project (they’ve already nailed down $2.75 million), they’ve been scrambling to get the latest Pleo prototype working well enough to show venture capitalists. They assemble in their “office” — the kitchen of John Sosoka, Ugobe’s CTO — where they have painstakingly been bringing Pleo to life. Two dinosaurs, one a skeletal model, one with its skin, sit on the kitchen table among a viper’s nest of cables and voltage meters.

The group is attempting to load one Pleo with an especially jumpy personality so that when an investor pets it, the toy will yelp in fear. “We’ll say, ‘Oh, he must not like you!’” Chung says. There’s just one problem: Pleo’s mouth isn’t opening and closing correctly. Slack-jawed, the creature looks vaguely brain-damaged.

“It seems like there’s a little mechanical interference right where the joint is,” muses one of the engineers. While a few programmers peer at a 3-D schematic of Pleo’s tightly packed gears, Chung goes into brainstorming mode. They don’t have time to rebuild the head, he figures, so a quick-and-dirty hack will have to do. “Maybe we shoot some graphite in there? Something to lubricate it? John’s good at doing surgery on this guy, right?”

Sosoka winces. A few weeks earlier, during a marathon 38-hour testing session, he had accidentally loosened a wire inside Pleo and set it ablaze. “You could see the fire through his skin,” he recalls.

Getting Pleo this far has entailed more than a few chewing-gum solutions. When Chung first got serious about creating Pleo three years ago, he began tinkering with rudimentary prototypes to find the fewest number of joints that would produce realistic dinosaur movement. He settled on five — four knee joints, one in each leg, attached to a spine that could bend at its midpoint.

In his garage workshop, Chung assembled a metal prototype and practiced controlling it remotely, like a puppet. A video of the model made its way to Bob Christopher, a startup veteran who had sold a voice-over-IP firm he cofounded and was then running a sport scooter company. Though the mock-up looked more like a collection of Meccano parts than a dinosaur — it didn’t even have a head or a tail — the movement was spot-on. As it crouched in mock fear and then reared up curiously on its hind legs, the toy sent shivers down Christopher’s back.

“It was like something absolutely alive,” Christopher says. “I told Caleb, this thing is going to change the world. We’re going to make the first robot that you have a serious emotional relationship to. We’re going to do robots as lifelike as Blade Runner!” He pauses. “Except in, ah, a good way.”

The trick was to make it affordable. Chung figured Pleo would need nine motors to power the legs and spine and five to control the head and tail. It would need 30 potentiometers to be capable of proprioception — a sense of where all its body parts were. And it would need a dozen sensors to detect external stimuli — someone touching a limb, say — and to “feel” its feet on the ground. It was a ton of gear and, worse, Chung didn’t want any of it to be visible from outside the bot. Nothing could break the illusion of realism.

It didn’t help that most species of dinosaur had tiny heads that, scaled down to toy size, wouldn’t accommodate all the infrared sensors, microphones, and speakers that needed to be crammed inside Pleo’s skull. Weeks of research led Chung to the Camarasaurus, a “chambered lizard” that had been blessed with a particularly big head and chest, perfect for concealing electronics. Chung learned that paleontologists in Colorado had discovered a complete skeleton of a one-week-old Camarasaurus; he began sketching models based on the measurements in their academic papers.

By late 2005, the team had cobbled together a demo Pleo with a skeletal frame and cheap gears and began the central task: producing lifelike movements. This is where Chung’s mime experience came in. With his sense of animal movement, he could perfectly intuit the sort of organic, loping actions that would make this dino seem real. He would act them out for the programmers and animators, essentially bringing Pleo to life by channeling the movements of a real-life dinosaur.

“You want it to walk,” Chung says, slouching over in a pantomime of a four-legged creature. “But then you realize you need it to walk in a sneaky way,” he continues, crouching lower and gingerly shifting forward, like a tiger. “And you have to get it to walk like it’s depressed or angry.” The robot’s movement can mirror its mood.

To speed the process of transforming Chung’s ideas into instructions for Pleo, the team developed a system that could turn motion into code. (Ugobe is filing a patent for it, so the company won’t reveal the precise mechanism.)

BY SUMMER, Chung and his developers had created more than 200 movements for various parts of Pleo’s body — “phonemes” of motion, as Chung calls them. There were astonishing successes: They could get Pleo to stand on diagonal pairs of legs — the front left and back right, for example — which is a major benchmark of balance for four-legged robots.

Yet some problems bedeviled them. They spent weeks puzzling over how to get Pleo to lie down on its side to sleep. Chung finally talked to a paleobiologist who argued it was a physiological issue: Real dinosaurs probably never laid down because, like elephants, their joints didn’t easily allow it. Instead, they would rest standing up or in a crouch, so Pleo “sleeps” in a sort of bent-kneed slump.

In February 2006, the designers hit another wall. Pleo was too loud, indicating that the gears weren’t meshing perfectly and thus were wasting energy. That would be fatal for battery life; they wanted Pleo to go several hours before needing a recharge. (Aibo lasts only an hour and a half.)

Ugobe hired Kleiss Gears, a gearing specialist. Using sophisticated computer modeling to analyze how much force each individual gear exerts on its neighbor, then tweaking the shape of the next gear to optimize the interaction, the firm overhauled Pleo’s system. “It grew the gears,” Chung says. When Kleiss handed back the redesigned gears, some of them looked so weird — more like petaled daisies than sharp-toothed mechanical parts — Chung wondered whether they would work. But once installed, they ran in nearly perfect silence.

A greater danger was that Pleo was steadily becoming more expensive — and its release date kept slipping. Chung and Christopher decided to take a calculated gamble. They realized it wouldn’t be possible to develop Pleo on the cheap and make it a Furby-like sensation. So they made a virtue of this vice: If it was going to be expensive, they would make it perfect. They would aim for an audience of early-adopter geeks, parents seeking to wow slightly older children, and (perhaps optimistically) teachers using it to demonstrate how dinosaurs lived.

By August, the price of Pleo had risen from $199 to $250. Worse, the shipping date slid from December to April 2007, meaning Ugobe would miss the crucial holiday buying spree, which can represent nearly 50 percent of the industry’s annual US sales. But since Ugobe plans to sell the majority of Pleos online, the hope is that Web customers might be more tolerant of delays for a coveted toy.

Observers are holding their breath. “Very tough,” says Jonathan Samet, an industry analyst with the trade magazine The Toy Book, of Ugobe’s strategy to bypass the holiday season. “Chung has a toy background, and Pleo is definitely getting a lot of press. But it’s not easy and not something I would ever do myself.”

WHEN FURBY CAME OUT, stories of its remarkable artificial intelligence were legion. People claimed their Furby had mastered names or spoken words in a foreign language. One man even said his Furby woke him up squawking when the house was on fire; its sensors had detected the sudden light.

But Furby wasn’t really so smart. “Furby couldn’t do any of that,” Chung says, laughing. It was programmed to make simple, random responses to stimuli and burble half-English, half-“Furbish” phrases whenever someone spoke to it. It didn’t learn at all. In reality, it merely relied on a quirk of human psychology: If something displays even a tiny bit of intelligence, people tend to attribute much more to it. “People fall in love with their Roombas, too,” Chung says of the vacuuming robots. “They dress them up in outfits.”

Pleo, in contrast, will grow and learn. Chung hands me a sheaf of papers that describe its life curve: It begins with the “hatchling” phase, a brief period when it’s first turned on that never recurs (unless you reboot your dinosaur). Following a period of infancy, it enters “puppy” mode, when it can be trained most easily to adopt new behaviors: If it makes a particular honking noise and you play with it, it will repeat that honk to get you to play again. (Chung notes that you can even abuse a Pleo and make it bipolar by pulling its legs and refusing to “feed” it.) A few weeks after coming alive, Pleo hits “adolescence,” during which it will mysteriously howl at the sky and sniff the air. After that, its voice drops an octave and it’s an adult. Its personality won’t change much after that unless you wipe its memory clean and start fresh — or hack it by loading custom-brewed personalities into its memory card slot, which Chung expects enterprising geeks will surely try.

Pleo’s day-to-day reactions to its environment are a result of its internal state. It monitors how “hungry,” “sleepy,” or “happy” it is, based on what’s happening around it. Then it responds with one of the movement phonemes: An angry honk if you pull the tail, a puzzled shake of the head if you cover Pleo with a handkerchief. When several possible movements are equally suitable, it will pick one randomly — which Sosoka hopes will keep Pleo from acting repetitively and predictably.

Last spring, while showing a prototype to a reporter at the San Francisco Chronicle, Pleo’s emotional system misfired and the servomotors generated two different bodily responses at the same time. The dinosaur began twitching violently as it flipped rapidly from one to the other several times a second. “Oh my God,” Ugobe’s PR manager screamed. “Pleo had a seizure!”

CHUNG’S VISION of lifelike robots stretches far beyond Pleo — indeed, it stretches far beyond toys. As Chung sees it, the world is going to have more and more service bots, so they ought to be physically and emotionally pleasing. They ought to be as lifelike — and as lovable — as possible.

In his workshop, Chung shows me an example of what he’s aiming for. It’s an experimental interactive chair he’s designing that “tries to understand you,” he says. A gorgeous, brushed-chrome affair, it’s driven by a set of hydraulic levers that fold it into several configurations. Chung is writing software that will enable the chair to analyze your body language as you approach it and configure itself to suit your mood. If you’re coming home exhausted after a long day at work, it will recline into an Adirondack style suitable for having a beer. If it’s a weekday morning and you’re holding a handful of papers, it will arrange itself into an upright work chair high enough to use with a drafting table.

Chung pulls out a manifesto he has written that lays out his design goals: “Objects that try and behave according to their nature,” it reads in neat block printing. “Objects with empathy. Objects with a purpose. Who will wake them up? Who will give them voice and action to their design? Who will write the interface to everything?”

The idea of making empathetic machines raises eyebrows, even among some roboticists. Sherry Turkle, an MIT professor and the author of several works on technology and identity, asks, “Should we be creating robots just to make people feel good? Should we be making artificial companions? Isn’t this a statement that we’ve given up on offering actual human companions?”

It’s a good point — though it’s also hard for me to worry about when I’m petting a Pleo that’s cooing and pushing its head into my hand, like a tiny dog eager to please. Maybe I’m just a sucker, but if this is a glimpse into the future of our machines, the future is going to be awfully charming.

Then Chung heads to his backyard to show me something even more adorable: a trio of tiny black Labrador puppies. (Chung’s daughter rescued a pregnant stray.)

“If you’re making artificial life-forms, you’ve got to keep real life-forms around, right?” he says as he scoops up one of the fuzzy dogs. It raises an impossibly cute, oversize paw and goofily sticks out its tongue. “The emotions, when you pick up these guys!” Chung says. “It’s a good reminder of the goal. We’re years and years away. But we’re trying. We’re trying.”


blog comments powered by Disqus

Search This Site


Bio:

I'm Clive Thompson, the author of Smarter Than You Think: How Technology is Changing Our Minds for the Better (Penguin Press). You can order the book now at Amazon, Barnes and Noble, Powells, Indiebound, or through your local bookstore! I'm also a contributing writer for the New York Times Magazine and a columnist for Wired magazine. Email is here or ping me via the antiquated form of AOL IM (pomeranian99).

More of Me

Twitter
Tumblr

Recent Comments

Collision Detection: A Blog by Clive Thompson