Arnulfo Zepedazepeda@siggraph.org
One of the most notorious advances in 3D animation has been the creation of characters, sometimes just humanoid and sometimes as realistic as possible which add a strong emotional value and induce us to believe the story being told.
Synthetic actors are humanoid actors totally computer generated. Day by day, animation become less acceptable if they lack synthetic 3D actors when they should be there.
In the present, in order to generate them, techniques such as keyframing (establishing key positions in specific moments of time), rotoscoping (which is a more elaborate technique of keyframing, in which you obtain real images with control points and try to converge the 3D models with them) and motion control (which obtains in an automatic fashion the key coordinates and angles frame by frame from a real actor). In the future we expect to use less such traditional techniques, and have the capacity of a real director to manage synthetic actors in real time through ver high level interfaces (as to ask him to walk in a certain happy fashion from a corner in a room to the door!).
Among their possible applications which are not obvious are the recreation of real actors, live or dead in any situation; scientific or artistic simulation (for transport, space, accidents and new tools) and biomedicine (structural problems of the (columna vertebral), visual and language disfunctions.
The history of synthetic actors has many interesting points, but it should be enough to mention that in 1971 Parker produces a crude representation of the head and face in the University of Utah, and three years later advances in good enough parametric models to produce a much more realistic face. A very important milestone happens in 1985, when the Film "Tony de Peltrie" uses for the first time facial animation techniques to tell a story.
"Sexy Robot" was created in 1985 by Robert Abel & Associates as a TV commercial imposes new standards for the movement of the human body (it introduces motion control). In 1988 "Tin Toy" is a winner of the first Oscar (as Best Animated Short Film) to a piece created entirely within a computer.
Other important milestones important to mention in the evolution of synthetic actors have been:
1988: First real time acting by a human character on computer in SIGGRAPH 88,"Mike the Talking Head"
1988: Daniel Thalmann and Nadia Magnenat-Thalmann create a Marilyn Monroe character technically impressive, but it starts to raise ethical and moral questions.
1989: In the film "The Abyss" there is a sequence where the watery pseudopod acquires a human face. This represents an important step for future synthetic characters.
1989: Lotta Desire, actress of "The Little Death" and "Virtually Yours" establishes new accomplishments.
1991: "Terminator II", movie which marks a milestone in the animation of synthetic actors mixed with live action bases its sequence in an indispensable way in the credibility of synthetic actors.
1993: In "Jurassic Park" dinosaurs, even without being human actors, rise again the realism (this time there are scenes which prove very difficult to know they were generated synthetically) and it is expected from other projects in ILM the creation of synthetic actors with the same technology.
In spite of all these achievements, however, there are still many problems in different phases of solution yet:
1) Natural movements generated by computer, in which each synthetic actor has to have his own 'personality' when he walks. This is very important since human vision is sensitive to small variations from an person to another from the frequency of the walk down to the smallest details. Parametric models have been employed to offer some solutions.
2) Pressing of objects with hands. Up to now FEA (finite element analysis) models have been employed for this processes. The typical steps to achieve it are:
3) Personality and facial emotions, are solved with modern techniques of skin scanning and with models and parametric colors similar to the ones used for "morphing".
4)Hair is a very complex object to model and render since there are more than 100,000 hairs over a typical human scalp. Solutions such as explicit models are very resource demanding since 100,000 cylinders would be used to model it! Other solutions offered today include the use of volumetric densities (3D textures) which besides color in a 3D space offer a kind of "direction field" to know where each hair is directed to. This works only for short strands of hair and for textures that look much more like plush than hair. The best solutions (even as they are difficult to animate) involve the use of particle systems and pixel diffusion of individual hairs in which once the scalp is rendered (with a Z buffer algorithm saving the depth information at each pixel) it proceeds to render each hair generated with dynamic programming algorithms and overlaying it on the original image. An important detail to observe is how hair interacts with itself and gives as a result areas o high and low density and of high and low brilliance. (not good for animation!)
5) Dresses: A complex problem because of how clothing fits the body (especially not tight clothing) and how when we walk it follows the body but with a movement which depends on the collisions with it and with itself, (for example, a skirt). It has been solved up to now using FEA.
6) Vision based animation: to direct an actor through rooms and situtations involving obstacles and avoid the specification by keyframing of the oath through which it has to move we can recurr to models with "artificial intelligence", generating views from the "actor camera"(or eyes) so that an expert system may decide to path to follow.
Research has been classified by some authors to the following categories:
1) Geometrical and physical (modeling and animation of the human structure)
2) Dependences on the environment (static objects, moving objects and other actors)
3) Event and objectbased behaviours, with and without constraints.
Among the videos shown, the following could be appreciated:
1. Eurithmy: of Michael Girard, achieves to create an expressive natural movement based on techniques of constarints optimization. Contrary to techniques such as keyframing and direct and inverse kinematics, this piece was created using employing inverse dynamics (by using final positions and velocities, and taking into account propesrties such as mass, force, moment of inertia, etc the movement is calculated) and resolving forces of the followinf kind:
.. . .
Q =S H q + SS C q q + g
i ij j ijk j k i
which are generalized forces, inertia tensors, and speed, velocity and generalized acceleration factors,
aiming to maximize the integral:
dJ = d3 Z(t)/dt3|^2
which is done to reduce the unnecesary reeling of the bodies extremities during the animation.
2. Still Walking: this video, as the next two, were created by Nadia Magnenat and Daniel Thalman. Starting from experimental biomechanical data and with a parametric model new realistic motion patterns are generated which imply different "personalities".
3. IAD: interface between the actor and the environment makes the actor conscious of the objects which surround him, making him able to grab them and to avoid colisons with obstacles.
4. Flashback: Animation of Marylin Monroe's dress as the wind hits it, taking into account self collisions and with Marylin's legs.
In brief, the description of a synthetic actor and the problems to create it are:
a) Its humanoid. There hasn't been yet a virtual Lassie. The creation of synthetic actors originates in the search of more realistic human models, limiting to the problems of the human body (as the generation of hair, facial expressions, articulations, human dynamics).
b) Personality. The distinguishing mark of an actor, virtual or real. The perosnality of a synthetic actor is not only given by his appearance, there are many other factors to consider: way to move, voice, reactions ans a long etc. With his personality, the synthetic actor will be original and unique and from it will depend his "role" in animations.
c) Able to operate on high level instructions. The ideal synthetic actor must be manageable in the same way as a real actor: telling him the actions that have to take place, without any need of controlling every step of each small motion necessary in order for the action to take place. It would be equivalent to just telling the synthetic actor to "Jump to your right, turn to the camara and show you are frightened", to which the actor will act performing all necessary intermediate steps.
Finally, its only necessary to add that synthetic actors are the one most exciting area in computer animation, and they will bring definite advances to the field. The techniques developed for their handling will be incorporated at some time to the modern computer animation software systems.
©Todos los derechos reservados, se prohibe la reproducción salvo permiso por escrito del autor.
©All rights reserved, any copying or duplication of this material is prohibited without written authorization by its author.
Return to Newsletters
Return to the Home Page of the Professional Group SIGGRAPH Mexico City