I can now blend animations together in my engine. For testing I'm just using a simple box with 5 sections with a bone per section.

During my searches for help with animation blending, I came across animation layering as well. I read a gamedev post where haegar wrote up some info on the differences but it was a bit above my head. Could someone please explain the difference between animation blending and animation layering with some easy examples?

My blending gave some interesting results. Animation A bends my box sections in one direction effectively creating a 'C' shape. Animation B twists the box along its length, effectively simulating torsion. Now if I give animation A 90% of the blend and animation B 10%, I get a 90% bend in the 'C' shape and only 10% torsion - just as expected, and it obviously works vice versa.

Now here's the thing, what if I wanted full torsion and full bend? Would that be where animation layering comes in? And how exactly would that work?

At the moment to do the animation blends I convert my matrices to quaternions, nlerp between them quaternions and lerp between the translations and then convert that back to a matrix. Not very efficient I know, but I'll refactor this later.

So looking at this from my game character's requirements, I would like him to have an idle animation, a jog and a run, so this would be a fairly simple blend between idle and jog and then jog and run, but why would I need layering? Would this be for something like running with the bottom half of the character and shooting with the top? (instead of animating arms as they'd be when running).

Thanks for any help or pointers

Thanks for the great explanation, I had a feeling it was that. So in a nutshell:

Animation blending is where all animation 'parts' (e.g. Walk anim, run anim) are pro-rata'd against each other using blend weights which add up to 1.

Animation layering is where you just add animation parts together with a weight for each and I assume it doesn't matter if the combined weight goes over 1, e.g. Layering run and walk would look odd but you can do it

So animation layering is the same as blending just without the combined weight normalisation?

And a use case:

My character is standing idle using the idle animation, breathing, swaying a bit etc. This is obviously just one animation, no blending or layering.

If the user leaves him in that pose for 10 or so seconds, to add a bit of eye candy, he maybe brushes something off his jacket. So I'd have an animation that only affects the top left half of his body, his head, shoulder, arms, etc. playing the idle animation, I can layer the jacket brushing anim over the top? If the jacket brushing anim has a weight of 1 and the top left half of his idle anim has a weight of 1, would that work?

The way I see it is that you'd use blending to fade nicely between animations and layering to combine animations together albeit to different parts of the skeleton.

So in the use case above, would I need to reduce the weight of the top left half of the idle anim down to 0 so the jacket brushing anim slots nicely in or can I just add them? I can't visualise what adding two anims together (making a combined weight of 2) would do - I'm guessing his hand in the jacket brushing anim wouldn't line up correctly with his chest because you've also got the idle anim in there. I guess I could add a priority and set the priority of the jacket brushing to higher than idle and restrict the combined weights to 1 meaning the top left half of the idle anim won't have any effect.

Is this about right?

[quote name='RobMaddison' timestamp='1332836675' post='4925577']
Animation layering is where you just add animation parts together with a weight for each and I assume it doesn't matter if the combined weight goes over 1, e.g. Layering run and walk would look odd but you can do it

Not exactly, layering and blending is more or less technically the same. The weights should always sum up to 1. It is more of an other term for combining animations.

An other example:
You have a walking character with his gun ready. Then he starts running while targeting his gun at an opponent in front of him. In this case we have 4 animations (walk,run,gun ready, gun aiming). We could divide the animation on two tracks or channels (movement and action). When our character starts running, the walk animation will blend over to the run animation, on the other hand, the gun ready animation will blend over to the gun aiming animation. Something like this:

movement_ani = blend_animation(walk,run,alpha)
action_ani = blend_animation(ready,aim,beta)

Alpha and beta are time dependent. Now you need to combine the movement and action animation by laying the action animation on-top of the movement animation with a bone-dependent weight, this is infact blending too.

final_ani = blend_animation(movement_ani, action_ani, bone_dependent_weight)


More advanced approaches are vertex animation overlays (i.e. facial animations not done with bones) or inverse-kinetics (i.e. pointing the characters arm always in the direction of the mouse).
[/quote]

I see, so when you do the third blend (the one between the movement and the action anim), you have per bone weights which essentially means the lower body will be fully (or almost fully) weighted towards the movement anim and the upper body will be weighted heavily toward the action anim...?

Thanks for all your help so far. I would have thought layering would be an excellent way of saving data - if only part of a skeleton needs to be animated, you only need key frames for that set of bones. I guess the conversion tool (between DCC and the engine) could work out that during the whole animation there's no movement of certain bones so just flag them accordingly so they're not part of the computation.

I think at this stage I just need to be able to animate skeletons, blending between poses and layering part-skeleton animations (eg the jacket scratching from above) on top. I also would like to be able to animate arbitrary elements, like the x element of a vector, etc..