A crowded sidewalk is a cacophony of unspoken yet unmistakable messages. A young woman’s “I feel sexy” walk, for instance, is instantly distinguishable from a biker dude’s “Don’t mess with me” stride. But getting computer-generated (CG) characters to reproduce physical attitudes like these is still an arcane craft. Animators must either eyeball characters’ movements in hundreds of hand-drawn “key frames,” with software interpolating the in-between moments, or cheat by using expensive motion-capture systems to digitize the behavior of real actors.
As a computer science graduate student at the University of Washington in the early 2000s, Karen Liu set out to find an easier method. Her article of faith: “There [had] to be some way, from our knowledge of physics and biomechanics, to distill the properties that create motion styles.”
Biomechanics researchers had long been analyzing the mechanical factors that affect the way people move. Simulating those factors, Liu thought, would yield CG characters that move more naturally. But the human body contains hundreds of interacting parts, and it was impractical to measure or even stipulate the values of parameters such as tension and elasticity for every muscle, tendon, and ligament. Working with advisor Zoran Popovic, Liu eventually showed that feeding just a handful of these values into animation software is enough to reproduce a distinctive motion such as a “happy walk” in a range of CG models, from people to penguins.
To establish her style parameters, Liu developed algorithms based on a single, simplifying assumption: that people naturally try to waste as little energy as possible when they move. Into these algorithms she feeds short segments of motion-capture data from subjects instructed to move in a certain way–to walk happily, for example. The software then reasons backward to guess the values of certain parameters, choosing those values that would have made the movements energy efficient.
Liu, who just joined the computer science faculty at Georgia Tech, is talking with major game makers and film studios about applying her algorithms to video games and animated films. She hopes the algorithms will help animators create CG humans that move more naturally than the robotically stiff characters in films like The Polar Express. “I think we’re really close,” she says.