Slashdot Mirror
Virtual Stuntmen Ready for Hollywood
Kerhop writes "Discovery Channel has an article about a new form of virtual actors in movies. In particular "Endorphin's virtual actors learn how to move and react independently, unlike most computerized characters now that depend on fixed databases containing animated clips". MSNBC also is featuring a news video (no direct link is available, stream must to be added to playlist). The featured software Endorphin is created by Natural Motion."
Forget movies...this would/could revolutionize gaming. Could you imagine actual thinking/reacting NPC characters that are not just set to a script. And if literally you can get fluid muscle movements... the graphics could be amazing. Okay, I'm done fantasizing. Regardless of how many times I see new "Revolutionary" technology it never seizes to amaze me.
June 26, 2003
The world's first virtual stuntmen, born out of an Oxford University zoologist's research into human motion, will make their debut next year in the film "Troy," according to a press release issued this week by the university.
I thought this site was supposed to be News for nerds. This story is over a year old.
Let's play video games with mailmanZERO
There are advantages to doing your own stunts. At least today the virtual actors are shown only from a distance, since facial features still don't quite look real enough (though even before Troy there were some lovely examples in Lord of the Rings. Those were keyframed rather than virtual actors.)
Being able to focus up-close on an actor doing a stunt gives a verisimilitude that the audience really appreciates; they believe that the character is in danger. But they can also detect the jump-cuts that usually surround a stunt (since a stunt is shot on its own, with lots of preparation beforehand, and the camera is immediately stopped so that everybody can be checked out.)
So when they're ready to seamlessly slip a virtual actor into the frame, then back to the human actor at the end, and make it look like a single shot, you'll really be thrilled. You will believe Brad Pitt can kick ass. (I once played Achilles myself and I loved Brad Pitt's work.)
What relation does this technology have to the Weta Digital's Massive program, which was used to create the characters for the war scenes in the Lord of the Rings movies? That program used sub-elements called Agents which could be tailored to give you certain randomated AI actions for characters. (Check the LOTR DVDs or google for Weta Massive for more info... I don't want to slashdot anyone in particular ;-) )
Many "stunts" are actually compositing and involve both real actors and real fire, just not in the same frame. Compositing can actually be done with classic film techniques, and does not require CG, though CG has become cheaper for most of them.
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
Within 10 years, someone will make a full-length, CGI feature film that's indistinguishable from "live, shot on film."
It may not be cost effective, but the cat will be out of the bag.
The actor's guilds will moan and groan. Actor's "trademark voices and mannerisms" will become protected intellectual property, with exemptions for spoofs and other "fair use." There will be court fights for awhile.
You won't see many new film actors after that point.
Porn will be among the early adopters go "all digital" for hygenic, financial, and other reasons. We'll have a lot of unemployed former porn stars.
On the plus side, the division between movies and games will blur. As someone else already hinted, games of the future will look "live" rather than animated. Sports games will "star" real NBA players that look as real as they do on TV. Movie-based games will "star" the same characters as in the movies, and will look just as real.
One potential downside to this technology that must, and will, be worked around:
Videotapes of crimes will become suspect. Unless you can prove beyond a reasonable doubt that the tape was NOT created in a computer, it will be inadmissable in court. You'll see security cameras equipped with "electronic seals" that stand up in court. These seals will say "this is what the camera saw and this is when it saw it."
Knowledge is how to play a game, intelligence is how to win, wisdom is knowing what game to play.
personally i think "outsourcing" is GREAT. If somebody is willing to do your job for half the pay, then you were probably getting over paid anyway.
Fuck you, man! Their cost of living is also lower. Indian programmers often have maids. I cannot control my cost of living. Plus, bidding low does not increase one's chances of getting a job here. I have tried that. It is such an uneven playing field that the ball automatically rolls into their goal. It is like being sentenced to unemployment simply because you are born in American. If it is a choice between programming and living in India or working as a greeter at Walmart and living in the US, I would chose the first. But, I don't have that choice simply because I was born here.
Plus, India does not have to be a parasite of the US to lift itself out of poverty. Why does everyone think that being a US corporate parasite is the only way out of poverty for those countries?
Table-ized A.I.
Or I should say we just had a demo. I work for a Digital Post Production house and we also use Massive for crowd simulation. Endorphin is an interesting technology which I believe demos very well but practical use may prove to be a bit harder.
The basic concept is intersting. Actors are "trained" using a neural network setup. The actions are not captured or keyframed, they are "learned" based on a fuzzy set of rules that allow the actor to adapt to its environment. These behaviors can then be combined to create a complete motion. For instance, in one of the demos you can have one actor tackle another. This is something you could easily motion capture, but the interesting bit happens when you change the direction of the tackle. The actor adapts and falls in a new way because the "tackle" is a behavior. The combination of dynamics and adaptive motion allow you to make changes without having to build lots of new in-betweens or blend shapes. It also allows you to set target poses, not target positions. For instance, in that tackle, if you want the actor to end in a Hesiman pose and then change the animation to a different position, the actor will finish in that pose in the new end position, not snap back to where you set the pose.
Unfortunately, this also has some big drawbacks. These behaviors that are taught to the nueral nets can only be built but the software's creator. You cannot create your own behaviors. I'm sure this type of thing will be opened up in the future, but for now you are stuck with some canned behaviors. Again, makes for great demos but I can easily see hitting the wall with that limitation in production.
Secondly, it only works on bipeds, or more specifically, their biped. Motion re-targeting will allow you to remap the motion to non-human bipeds, but you are still limited to bipeds with human like bio-mechanics.
I also wonder about low level noodle-ability. In animation production, it doesn't matter if it is physically correct, it matters if it looks good and the director is happy. That usually has very little to do with reality (i.e. how interesting would the Matrix have been if people moved in real-world way instead of an interestingly choreographed ballet-fight?) I wonder how adaptable such a sytem is.
This product has a different market than Massive. Massive includes a more simplified fuzzy brain system that allows your motion captured creatures to adapt to their environments or even interact with dynamics. Endorphin is more about syntesizing motion from neural net behaviors. (That is a gross simplification, but hopefully gives some context.)
I like the idea of where this is going, but I think it has to mature a bit more before it is really useful to us.