Associate professor in the Rennes 2 University LPBEM laboratory (1), Franck Multon is currently an invited researcher with the Irisa Bunraku project (2). He presented his research on computer generated images, animation, modeling and simulation during his HDR defence (3), on December 8, 2006.

Computerization techniques are used in several application fields in order to animate human-like figures, referred to as avatars or humanoids. Usually, they do not produce enough realistic trajectories. But things are changing. “Until now, computer simulation was based upon bone angle measurements. The process was hard because it had to deal with so many motion possibilities”, says Multon. With 206 bones, 650 muscles, tens of rotations, intertwined lengths of movable joints, connective tendons and ligaments, the human body is a complex machine driven by a string of laws such as gravity, mass, force, or inertia. Now, factor in a crumb of psychology and any human gait can turn into everything from flabby shuffling to flippant loping, adding complex niceties to the matter. Modelling and simulating the movements of such a mechanism is therefore rather… “challenging”.

The new approach takes a completely different strategy. It is based on a simplified skeleton with normalized global postural information and kinematic constraints that are intrinsically linked to the motion, such as ensuring foot-contact with the ground or reaching targets for grasping motions. The magic of it is “a phenomenal reduction of calculation time, Multon sums-up. It runs 100 times faster.

Second originality: “the method is interactive. Many researchers need to know the full animation scenario in advance. This creates a problem: it can’t deal with unexpected events, like... changing the trajectory of an object that has to be grasped by the humanoid, for instance. In our system, we don’t need foreknowledge.”  A dramatic illustration of this dynamic adaptation is a video animation of avatars dancing on a waving floor. The motion looks natural, smooth and flawless. Such quality wasn’t available previously as most 3D animators provide their virtual character with specific size, weight and so on. Resulting drawback: the motions can hardly be animated on characters having different morphologies. On the opposite, Bunraku’s approach is based upon a morphology-independent representation of the skeleton. Thus, it allows the animation on the layman computer of hundreds of different characters, from lumberjack to ballerina, while at the moment most research centers can hardly animate more than one character at the time.