Hi,
I'm also an indie game developer, and i wanted to add procedural/dynamic physics based animations to my game engine, mainly because i'm an aweful animator.
anyway, I've been looking into the work done by Petros Faloutsos and a few other who i've lost the links to there pages...
anyway, in my searches i've stumbled accross wolfire, and lugaru, and after days of searching i've found a similar application which i'm presuming (and praying) doesnt require insane quantities of data input or a motion capture machine.
although i'm in the early stages of development (currently optimising IK algorithms..) I'm moving towards this type of animation... and i want to know everything about it! or at least some information.
from what i can guess sofar is that theres some controller AI FSM that chooses actions (walk run jump) and those actions from the youtube videos are made with learning algorithms on a model that has physics applied to it.
I'm interested in what kind of learning algorithm is used (genetic & darwinian AI, neural nets?), whats input into the learning algorithm, presumably the learning algorithm provides FK for a certain task, and then IK is applied to the extremeties(hands/feet) to make good contact.
are there any papers/theses that i can read that were used to develop these
agh, so many questions, though i'm most interested in the learning algorithm. =]
sorry if this is out of place, or considered rude. I've just spent days at this with minimal success, I was very excited to stumble upon wolfire at 2am =p
thanks allot,
Dead.Rabit
procedural animations from lugaru 2
Re: procedural animations from lugaru 2
Unfortunately, I don't think you're going to get what you want here. It sounds to me like you're after a Spore-like mechanism for creatures to automatically figure out how they should move. From what I've inferred from statements by the Wolfire devs, though, they don't have anything like that. In this case, "procedural animation" refers to programmatically combining animations (e.g. combine running with attacking), making animations aware of the terrain (so you don't have intersections when your rabbit runs uphill, etc.), allowing for smooth turning as your rabbit runs in circles, and so on. The individual animations still have to be created; it's just that they can be cleanly extended to a variety of situations and work well with each other.
Of course, I'm not a dev myself, and I could be completely wrong about how Overgrowth's animations will work. But to me, it sounds like a lot less work to make animations directly than it is to make a learning algorithm that comes up with the results you want, especially when you're talking about highly specific motions like those you find in martial arts, where you can't depend on your learning algorithm doing it right. That's not to say that making animations is easy, of course.
In any event, I wish you luck with your project.
Of course, I'm not a dev myself, and I could be completely wrong about how Overgrowth's animations will work. But to me, it sounds like a lot less work to make animations directly than it is to make a learning algorithm that comes up with the results you want, especially when you're talking about highly specific motions like those you find in martial arts, where you can't depend on your learning algorithm doing it right. That's not to say that making animations is easy, of course.
In any event, I wish you luck with your project.
Re: procedural animations from lugaru 2
Heyya,
Actually i was hoping to get results very similar to lugaru, I would have to FK (well im looking into alternatives but) for some actions. however in terms of a physics model keeping its own balance under gravity and wind, being able to walk and turn in any direction, roll or jump in any direction, would be impossible to FK.
also with regard to slopes and styles of motion.
I wish to teach my charachters how to move, to take a protective step to stop themselves falling, steppingin the correct direction to to stop falling.
well, thats the eventual goal of th engine, well, i've got to put in collisions and the underlying physics engine first hopefully by next week, then i want to work on my AI for controling player charachter motion.
thanks
D.R
Actually i was hoping to get results very similar to lugaru, I would have to FK (well im looking into alternatives but) for some actions. however in terms of a physics model keeping its own balance under gravity and wind, being able to walk and turn in any direction, roll or jump in any direction, would be impossible to FK.
also with regard to slopes and styles of motion.
I wish to teach my charachters how to move, to take a protective step to stop themselves falling, steppingin the correct direction to to stop falling.
well, thats the eventual goal of th engine, well, i've got to put in collisions and the underlying physics engine first hopefully by next week, then i want to work on my AI for controling player charachter motion.
thanks
D.R
Re: procedural animations from lugaru 2
The animation in Overgrowth does not use machine learning techniques at all. I have experimented with that in the past and just found it to be not practical for what we are doing. It is all about automating traditional animation practices, like anticipation/response and squash and stretch. I am planning to make a video showing the step-by-step development of it within the next couple weeks
Re: procedural animations from lugaru 2
If thats the case do you essentially then work with keyframes and automate the tweening with the physics procedure?
Re: procedural animations from lugaru 2
I'll look forward to that video, =].
D.R
D.R