Chapter 94 16.3 Robots without bodies

At Crosall Graphics Studio in San Francisco's industrial district, Brad de Graff is working on replicating human behavior.Closar is a little-known special effects studio that has been behind the scenes of many famous TV animated commercials.Closar also produced a pioneering animated series called "Moving TV" for MTV.The cartoons are starred by rough-and-tumble characters — down-and-out rag dolls on motorcycles, lifelike animated cutouts, and bad boys with puppies and pig brains, to name a few. Graff's studio is cramped in a renovated warehouse.In several large dimly lit rooms, two dozen huge monitors flickered.This is an animation studio from the 1990s.The computers, all powerful graphics workstations made by Silicon Graphics, flashed images of various stages of the project, including a fully computerized bust of rock star Peter Gabriel.Gabriel's head and face were scanned and digitized by a computer, and then stitched together into a virtual Gabriel to replace his real body in the music video.These things can be done in a recording studio or on a dance floor, who bothers to dance in front of a camera?I watched an animator fiddle with this virtual star.She was about to lift Gabriel's jaw by dragging the cursor so that his mouth would close. "Oops." She exclaimed, she had moved a little too far, and Gabriel's lower lip was raised too high through his nose, making an ugly grimace.

I went to Graff's studio to meet Moxie: the first fully computerized animated character.On the monitor, Moxie looks like a cartoon dog.He has a big nose, a gnawed ear, white-gloved hands, and "rubber tube" arms.He also has a very funny voice.At that time, his movements hadn't been drawn yet.The motions are extracted from the motions of a human actor.In one corner of the room, there is a homemade virtual reality device "Valdo".The so-called Waldo (named after an old science fiction character) is a device that allows people to manipulate puppets from a distance.The first computer animation done in this way was the experimental Komi Frog, drawn using a palm-sized Waldo device.And Moxie is a virtual character with a complete body, a virtual puppet.

When the animators wanted Moxie to dance, he would put on a yellow helmet.There is a small stick on the top of the helmet that is secured by tape, and the end of the stick is a position sensor.The animators then strapped sensors to the shoulders and crotch, before picking up two giant cartoon hands cut out of foam board—gloves, actually.He's waving those two hands as he dances—there's also a position sensor on them.So Moxie, the cartoon dog, also danced in step in his weird mahogany house on the screen. Moxie's best trick is automatic lip sync.Feed the recorded speech into an algorithm that figures out how Moxie's lips should move and then moves them.The masters of the studio always make Moxi say all kinds of annoying things in other people's voices.In fact, there are many ways to make Moxi move.Spinning a dial, typing a command, moving a cursor, or even algorithmically generating some sort of autonomous behavior can make Moxie move.

What Graff and other animators want to do next is to give characters like Moxie certain basic movements -- getting up, getting down, carrying weights -- that can be combined into coherent and realistic movements.Then it can be applied to complex human characters. Given enough time, today's computers are barely capable of computing human actions.But to do calculations in real time, like your body does in real life, the simulation is almost impossible to calculate.The human body has about 200 points of motion.The number of gestures that these 200 movement points can make is basically an astronomical figure.Just the real-time action of picking one's nose requires more calculations than the mainframe computers we have today.

And the complexity of human movement does not stop there, because every posture of the body can be achieved through many different ways.When I put my foot in the shoe, I have to guide the leg through hundreds of movements of the calf, foot and toes to complete the entire movement precisely.In fact, the sequence of motions my limbs perform in walking is so complex that there is room for a million different ways of doing it.Usually, acquaintances recognize me from a hundred feet away without looking at my face, simply because I use my dominant leg muscles involuntarily as I walk.It is very difficult to imitate other people's movement combinations.

Researchers trying to make artificial figures mimic human motion quickly realized what the animators who made Bugs Bunny and Piggy had known for a long time: Certain joint sequences appear more "natural" than others in terms of motion. ".When Bugs Bunny reaches for a carrot, the path of its arm reaching for the carrot is more like a human arm movement (of course, Bugs Bunny’s behavior is not imitated from a rabbit, but a human), and the timing of the movements of each part is also very close. relation.Even if an animated figure moves in the correct sequence for a human, if the relative speed of its arms and legs can't keep up with the rhythm, it will still appear mechanical.The human brain can easily spot such fakes.Therefore, timing is another complex aspect of action.

Early attempts to create artificial motion forced engineers to study animal behavior.To build a multi-legged vehicle that can roam Mars, researchers are studying insects, not to learn how to make a leg, but to figure out how insects coordinate the movements of their six legs in real time. In Apple's lab, I once watched a computer graphics expert play a video of a cat walking over and over to break down its movements.The tape, along with a stack of scientific papers on the instinctive reflexes of a cat's limbs, would help him refine the cat's walking style.He then intends to implant this style into a computerized virtual cat.His ultimate goal is to extract some general quadrupedal locomotion pattern that can be adapted for dogs, leopards, lions, or whatever.He was not at all concerned with the shape of the animals; his models were rough-and-tumble figures.His concern is how to organize the complex movements of the legs, ankles, and feet.

David Selcher of the MIT Media Lab led a group of graduate students to develop a thick-lined figure that can walk "by itself" on uneven ground.The images are simple - one line segment for the torso, and four line segments for the four legs attached to the torso.The students set a direction for this "little creature", and it moves its steps to find out where is high and where is low, and then adjusts the length of its steps accordingly to move forward.The result is a realistic image of a creature walking through rough terrain.Unlike the "Beeping Bird" animation we have seen, in this film, when the animal moves which leg is not determined by humans.In a sense, the character made the decision for himself.Selcher's team later added six-legged self-propelled "little creatures" to their world, and even created a two-legged creature that would wander around the valley and come back.

Selcher's students also assembled a walking cartoon called Lemonhead.Lemonhead's walk is more realistic and complex than those linear figures, because its movement requires more body parts and joint support.It can maneuver around obstacles such as lying tree trunks very realistically.Lemonhead inspired another student in Selcher's lab, Steven Strassman, to see how far he could go in designing behavior libraries.The basic idea is to give a generic character like Lemonhead a "scrapbook" of actions and poses.Want to blow your nose?OK, there's a whole disc of action to choose from.

Strassman wanted to direct the characters in plain, understandable English.As long as you tell it what to do, it will find a set of suitable materials from the "behavior library", and then compose them into a reasonable action in the correct order.For example, if you tell it to stand up, it will know that it should remove its feet from under the chair first. "Look," Strassmann reminded me before the demo began, "this guy can't write a sonata, but he can sit on a chair." Strassman uses two characters, one named John and one named Mary.The story takes place in an ordinary room; the point of view is looking down from a certain angle of the ceiling-this is somewhat of a divine looking down.Strassmann called this "tabletop theatre."The couple quarrels from time to time during the play.And Strassman is going to have a scene where the two break up this time.Here's what he typed: "In this scene, John lost his temper. He roughly handed the book to Mary, but she didn't take it. He dropped the book on the table. While John was staring at Mary, Mary stands up." He hit play when he finished writing.

The computer thinks for a few seconds, and then the characters on the screen start acting.John frowned, and the way he handed the book was very stiff; he clenched his fists.Suddenly Mary stood up.Finish.Nothing fancy, and their actions don't look very human.It's not easy to catch those fleeting poses because their movements don't capture the viewer's attention.The people watching didn't have any sense of participation, they just knew that in this room, two people interacted according to the script designed by God. "I'm obsessive as a director," Strassman said. "If I don't like how a scene looks, I tell them to do it again." So, he typed in an alternate scenario: "In this scene, John feels sad. He holds the book in his left hand. He gently hands the book to the Mary, but she politely declined." The characters played the script again. The difficulty lies in the subtleties of the characters' movements. "The way we pick up a phone is going to be different than the way we pick up a dead mouse," Strassman said. "I can store different hand movements, but the hard part is not knowing what governs. These actions. Where did the department in charge of these choices come from?" Sercher and his colleague Michael McKenna took a page from the rough-and-tumble figure and the lemonhead, and set out to embellish the six-legged creature, turning it into a vicious, metallic chrome cockroach, making the little insect the most famous of all time. The star of one of the weirdest computer-animated films ever made.They jokingly called the short "Grinning Evil Reaper."The five-minute film focuses on a giant metal bug from outer space that invades Earth and destroys a city.Although this story is boring, the protagonist of the story, the six-legged monster, is the world's first "little living creature" - an internally driven artificial animal. As this monstrous chrome cockroach crawls down the street, it acts 'free'.The programmer tells it to "walk through the building," and the virtual cockroach in the computer figures out how its legs should move, what angle its torso should turn, and writhes its way across a building. 5 storey brick house.What the programmer gave the big bug was only the target of action, not the instruction of action.When coming down from upstairs, an artificial gravity will form a kind of traction for this huge robotic cockroach, causing it to fall down.When it falls, simulated gravity and simulated surface friction make its legs bounce and slide like they would in reality.This roach doesn't need its directors to struggle with its cumbersome legwork to get it right. People are experimenting with autonomous avatars a step further: take the bottom-up behavioral engine of a giant cockroach, wrap it in the glamorous shell of a Jurassic dinosaur, and you have a digital movie actor.Wind up the actor, give it plenty of computer cycles, and direct it as you would a live actor.Programmers just give it some general-purpose instructions—like “go get something to eat”—and it will figure out how to coordinate its own limbs to do so. Of course, dreaming is not so easy.Movement is only one aspect of action.In addition to moving, simulated creatures must find their way, express emotions, and respond.To create creatures that do more than walk, animators (and robotics researchers as well) need to find ways to breed all types of natural behavior.