Theory & Discussion: Combining two or more animations.

Started by
2 comments, last by JoeJ 3 years, 7 months ago

Hello,

I've always had a question about animation: is it possible to make an animation and then do another and try to combine the two to create a totally new unique animation with characteristics of both? And would it be possible to do this process with more than two animations only?

Would this have some kind of technology like machine-learning involved? And how would that apply to procedural animations?

I don't know if there are already games in the industry with something like that, but it is sometimes really difficult to search for something very specific on Google, so I came here to ask

Thank you in advance, and feel free to discuss about it and give your opinions ❤️
(just remember that I am an extremely beginner and just like to do experiments of all kinds on Unity)

Emanuel Messias, R.d.SBrazil, São Paulo
Advertisement

Sorry, I don't think I specified exactly

This video was actually useful because of the physics topics, but I already know about interpolation, layers and a little bit about animation line, so it was really useful for me. Thanks for that anyway (also, Unity has something similar to all these tools).

I will try to specify with an example:
Your character is very angry and you are executing the punch animation, I thought of combining a normal punch animation with some animations that express that the character is angry, so your punch will be more "aggressive".

Of course, this can be simulated with other techniques also presented in this video but I literally thought about the possibility of literally "combining" things and not just interpolating, I thought there was already an ML algorithm capable of doing this (the only method I could think of for this idea was machine-learning), but I researched for months and I didn't find anything, anyway, it's a very complicated thing to find any results on google (even more for me).

With so much difficulty in finding any results I am thinking that nobody thought or did something like that, so I say that it is a theory, but also an idea that I really wanted to implement in some project.
(I just don't know exactly how, because I have no other reference other than my imagination)

And I thought this would be a very good topic to discuss with other people who has more knowledge and experience.

Emanuel Messias, R.d.SBrazil, São Paulo

There is a lot of research going on, also based on ML. this is the first thing i've found using google.

But your question seems very broad. Some points:

Interpolation alone is not totally trivial, as we lack a simple way to mix 3D rotations. (In practice, linear quaternion interpolation should do, but just to mention.)

There exist many approaches to do get procedural character movement. Some of them including or entirely based on dynamics simulation. Some require existing data (like most ML methods that learn from motion capture), some don't.

I have some exerience with making a walking ragdoll based on dynamics simulation. I think that's very promising, but i also think ML can be combined with this to handle more complex and fuzzy actions than basic locomotion.

I assume the industry will move into this direction, but maybe fully ML based.

This topic is closed to new replies.

Advertisement