How to make a fat man jiggle in 3D animation

For cartoon characters like the hapless Homer Simpson, being fat is often a matter of pride.

But for 3D animators who have to model body fat - and the way it jiggles about as a rotund character moves - it's a major problem: doing it convincingly requires large amounts of computer processing power and expensive rendering time.

Computer scientists Junggon Kim and Nancy Pollard at Carnegie Mellon University in Pittsburgh, Pennsylvania, say that instead of simply modelling a 3D cartoon character as a deformable human shape, they should additionally build a separate skeleton into the body - and model the way the deformable material (flesh and fat) behaves as the skeleton moves it.

While skeletons are widely used to animate 3D characters, driven by motion captured from a human, they have rarely interacted with the deformable material that surrounds them. Kim and Pollard's innovation was working out an algorithm that can do that with the fewest possible processing cycles - and so the shortest rendering time, making it more economically feasible for animation studios.

Their 3D graphics algorithm attaches the volume pixels ("voxels") of the deformable body to the skeleton. "So when the skeleton moves, the fixed body points also move along the bones and this leads to the passive jiggling movement of the deformable body," Kim says. But they have also geared their algorithm for processing on parallel graphics processor units (GPUs) which, in the case of the 'Fatman' animation above, sped rendering by 10 times.

"Normally, this is a very time-consuming task because of the massive computation required to capture dynamic deformation of an elastic body," says Kim. "Our technique boosts the simulation speed by orders of magnitude while preserving the quality of the simulated motion." Their work also applies to nonhuman forms - you can see how they have applied it to the animation of a trapped starfish escaping from a box (like they do) on this page, where they reveal how these animations can be controlled. 

http://www.newscientist.com/blogs/onepercent/computing/