Automated animation

I saw that video :

[Blocked Image: http://youtu.be/pgaEE27nsQw]

@Zarsky I think the two are very different , since IK rig is based on classic skeletal animation, while what I've shown has been "learned" by the machine. They've run an algorithm and at approximately 1000 tests, the skeleton could move itself like it should, without any human telling him how to do it.

Quote from Zarsky

4 hours ago, Zarsky said: Oh I have to take a closer look. Sounds interesting....wonder when all the NPCs will be controlled by the cloud running complex simulations on AI, behavior and movement. Games will never be the same after that.

Why on the cloud?

Quote from Zarsky

1 hour ago, Zarsky said: I guess I was just thinking about those self learning neural networks.....but yeah, things are advancing at a rapid pace, so the required computing power might be available soon enough offline.

The idea is that the exploration part requires computing power, but once it's done, the exploiting part isn't that costly, and can be run with a decent hardware

Quote from Logic

12 hours ago, Logic said: I guess Automated Animation is one way to put it, but I don't believe this is developed to do any actual animation work.

Anyways, I saw this video a while back and I loved it! It would be interesting to see a game that makes evolution a gameplay element and puts you in a creator role Spore won't stand a chance!

How would you have called it? It's animation made automatically

I wondered if it was that's why I posted that, but apparently people in game studios don't want to tell if they use that...

A non-scripted evolution? That would be cool yeah, however that would mean using GPU's for that, so you'd have to have a lot of things precomputed anyway, so you don't overuse the GPU for that

Quote from Logic

23 minutes ago, Logic said: I would have called it Evolutionary AI. The main goal is simply to travel as far as possible using muscles in the different bodies. It doesn't matter if the creature flops around like an idiot or walks upright as long as it finds the most efficient way to traverse the landscape. The animation is just a bi-product that comes with the solution to getting from one position to another.

Again, I don't believe it's actually used for animation, but would be really cool if it did!

I don't think game studios are really scientific enough for it to be used. I mean, I know Ubisoft has a machine learning division, but no-one like them so no one else is doing it as well

Quote from Logic

4 minutes ago, Logic said: I don't think it's worth it en the end. It's probably easier to do motion capture rather than developing AI that might not even live up to expectation. It does have potential however.

External Content www.youtube.com

Content embedded from external sources will not be displayed without your consent.

Display content once Display all external content

Through the activation of external content, you agree that personal data may be transferred to third party platforms. We have provided more information on this in our privacy policy.

The closest thing to this is probably the implementation of the Euthoria engine in GTA IV. Ragdolls weren't really learning anything, but had properities that would make them react in different ways, depending on the situation. Interesting stuff

That may look a bit like what I've shown though

It may be using it! Well there a few similarities about those, when you look at how they can be affected by the environment and keep still, that's the real advantage of machine learning based animation, they are flexible and unique each time, less memory space for a diversity of animations.

I think you answered my question, it is used apparently

@spence from what I understand, it doesn't have to be translated. That optimum walking animation is learned by the machine, meaning that the next iteration it will take into account what made it fail, you don't have to simulate all the 900 iterations before that one.

A friend of mine explained me, in deep learning algorithms, there is a discovery phase, where the algorithm is exploring possibilities, assigning them a reward value, based on an objective stability criteria. That's the process showed in the video. After that there is an exploitation phase, where it uses the results of the exploration phase to make the best calls possible. It therefore uses a kind of memory to do so. It is of course more expensive than using standard predefined animations, but it isn't extremely demanding either and a reasonably good computer should handle a dozen of calculated animations. (It might get an overhead with thousands of people to animate at once, but that's not something which should happen in a video game, because with that many moving pieces you'd have to simplify things, but that wouldn't change most of what's being rendered anyway.