Slashdot Mirror
UCLA Adds Physics to Prat-falls
BaltoAaron writes "CNN.com is reporting on Petros Faloutsos , a UCLA scientist, that has developed a program that creates animation based almost solely on physics. Faloutsos "believes his animation program will one day allow virtual stunt artists to replace their flesh-and-blood counterparts in performing otherwise deadly feats of derring-do." "It's the Holy Grail of character animation. Everybody wants to do it, but there's not a whole lot of it out there right now.""
DANCE is a portable, open, plug-in based, object-oriented software package for physics-based character animation. It runs on Linux, Irix, Windows 98/NT and is being ported to MacOS by Joe Laszlo. One of its goals is to provide researchers with a common platform where they can test their control methods and share their results. In addition, it provides the common, yet complex functionality that everyone needs in a physics-based animation system, allowing researchers to concentrate on their research work. Dance has been used for a variety of physics-based applications that include biomechanics modelling and composeable controllers. For more information, please contact the authors.
It's available to download and play with!
There has been research in this area being done for years, much of it presented at SIGGRAPH. There are techniques to animate characters through intricate plots just by specifying behavioral charactics, techniques to apply motion dynamics to characters of significantly different shape, and even "video puppetry" that allows images to self-animate in response to speech. All are a number of years old. All were hailed as holy grails. This just seems to be a case of CNN finally noticing.
At last year's SIGGRAPH, everyone already knew about polynomial textures, because there had been a news story about it. To me, though, the highlight of the show was that it is now possible to walk around with an uncalibrated, handheld camera, and completely automatically get a decent 3-D model out of it (textured, of course). No news story about that.
Kinda, but its always a pretty significant simplification. Only recently games have started adding things like Inverse Kinimatics(IK) in order to create motion sequences that are not just replays of motion capture.
These days a game programming text looks like an abridged edition of a scientific modelling text.
A big part of the trick is to have a realistic model of the human body. There are hundreds of joints of several varieties and many muscles controlled by the worlds most complex single entity(the brain). This makes it very hard to come up with a 'first principles' model. This is why most animation packages today (AFAIK) model the human body as a series of rigid parts(bones) connected by springs(muscles) with control points that the animator can use(the brain).
Kevin
Thoughts on tech, Software Engineering, and stuff
Years and years...
http://www.animats.com/
Physics-based animation has been a hot topic in computer graphics for a decade. SIGGRAPH made a major award to Prof. Andy Witkin of Carnige Mellon in 2001 for major progress in this field. This involves anmal motions, objects colliding, objects shattering (e.g. Phantom Menace) and so on.
Every computer generted graphic movie thus far has failed
Uh... Toy Story I/II, Bugs Life, Monster's Inc, Shrek, Antz (which I think sucked but did good business)... I wouldn't exactly call them failures.
The researcher performed this work for his
Ph.D. thesis at Toronto. Though he is
most likely continuing the line of research
as a professor, the article is about his
thesis work.
Sorry, but you are not very clued in on the
research. Petros did not make a physical
simulation of a human walking. That had been
done many years earlier. Researchers at
Georgia Tech [Hodgins, et al.] and U Penn
[Badler, et al.] have focused on simulated
humans since the early 90's, simulating motions
from running to bicycling to diving.
Petros's work was on integrating these motions
together: so a character could walk, trip,
dive, land, roll, and stand back up again.
He used support vector machines to learn
the domains of acceptable pre- and post-
conditions of different movements and plan
the transitions.
When you use forces, as you suggest in
the second paragraph, you are not doing
forward/inverse kinematics, but rather
forward/inverse dynamics, a much harder
problem.
And things like trees and jello behave
passively, that is they don't produce any
forces on their own from muscles, motors, etc.
My guess is what you are referring to in the
first paragraph is simple spring-mass systems.
Modal analysis can be used to obtain more
accurate deformations for things like trees.
But if you want to simulate humans, you need
to model the human's muscles as well if the
human is anything but limp. The interaction
can be very complex (especially given closed
loop situations such as two legs on the ground).
Faloutsos' work is not actually focused on the physics models, but on the control programs for the virtual actors. This allows dynamic, force-based animation (as opposed to kinematic, position-based animation). Each model has a set of controllers for various tasks like walking, running, jumping forward, moving from a prone position to a standing position, etc. Each controller knows its "competencies" -- the conditions under which it can successfully guide the model. These are used to hand off control from one controller to the next as the model goes through a complex motion or reacts to external forces.
The sample movies that Faloutsos showed were mostly unscripted. They would start with a model in a simple standing state, which would then respond to user-controlled forces like pushing or throwing simulated balls at the model from various angles. Various balance-recovery controllers would take over depending on how the model was displaced; if none of them were succesful then the model would fall down, and then use one of its controllers for returning to a standing position. All of this appeared incredibly realistic and human.
Also, as another poster noted, DANCE is available under a "free for non-commercial use" license (not free under the FSF or Debian definitions, but a good deal in my opinion). He encouraged us to try it out, explaining that research like his has suffered from a lack of common infrastructure, leading to a lot of reinvented wheels. He expressed hope that the DANCE framework would allow more innovative research with less duplicated work.
http://www.cs.ucla.edu/~pfal/animations.html