Unexpected rotation animation blending result (quaternion slerp vs animation blending)

[nsf]

Godot version

4.0 alpha 6

System information

Linux, Vulkan

Issue description

I have a simple blend tree setup where I blend two animations. Each animation consists of a single quaternion keyframe. Thus it's not even an animation per se. Of course this is a minimally reproducable example, in real use case I blend between proper two animations.

So, basically, I blend between two quaternions and I expect the result of such blend operation to be equivalent to "slerp" between two quaternions, in particular the shortest rotation path needs to be taken, however animation blending gives a different result. Here are the two videos of it.

Quaternion slerp using tweens and a script (this is what I expect):

slerp.mp4

And this is the blending between two animations using AnimationTree:

blend.mp4

I've seen some recent pull requests in this area, where code does some quaternion magic to make it "right", e.g. #57675, might be related.

Perhaps this small use case might be helpful.

Steps to reproduce

Run the test project, by default it will play the tween animation and you can inspect the AnimationTree node in the editor by activating it first.

Minimal reproduction project

I've pushed the test project to github: https://github.com/nsf/GodotSlerpTest

[nsf]

Additional note: it works as expected in 4.0 alpha 4

[TokageItLab]

@nsf I think maybe it has been already solved by #60093 and #60235. So If you are using ValueTrack or Rotation3DTrack with Node3D, you need RESET track to determine the direction of rotation; if you are using Rotation3DTrack with Skeleton3D::Bone (I guess above video is not so), you need set the correct Rest. I know this thing needs to be added to the documentation later.

If you can build engine, please test it, or wait next alpha.

[TokageItLab]

I tested, probably it seems fine with RESET.

2022-04-16.22.56.10-1.mov

GodotSlerpTest-main_fixed.zip

[nsf]

Yes, I just compiled it from git and I can confirm it works fine with RESET track. Thanks.