Hi, this is an extremely important domain for Organic Rigging.
I haven’t watched it yet, so I have no idea if that’s a Blender vanilla thing or if that is a simple Setup to achieve from scratch.
The main problem revolves around the brutal effects (“Artifacts”) of conventional CG Skinning methods/algorithms for more natural/organic/realistic Deforms, usually caused by Rotations on whichever Axis.
In Blender at least, I know there exist only 1 (one) possible Rest Position at a time; the Rest Position here I mean, is the 3D Model’s Mesh on Edit Mode, but more rigorously, its Skeleton/Armature Object set to Rest Position: Binding Pose State.
Shape Keys, although they create ‘Mesh’ Keys or Morphs, only the “Basis” Shape Key represents the true Rest Position. All of the other Shape Keys they Morph the Mesh, but they bypass or ignore the effects of the Skinning method/algorithm, and vice-versa. So that Deforms (from Skinning method or Armature Deform) and Morphs (from Shape Keys), they are independent processes. But that you probably already know.
If we could have multiple Rest Positions on a 3D Model in Blender, then that would probably mean it would make/force some sort of “Interpolation” between, say, 2 radically different Rest Positions, calculations made by a new ‘composed’ Skinning method/algorithm (as if the 3D Model was Rigged twice-in-one in this example; so we could have for example a 3D Character with a T-Pose Rest Position, and another Rest Position with a highly Bent Limbs in a certain more useful Direction of Bending; that wouldn’t be sufficient of course because certain Limbs have all sorts of Bending Orientations, including in Opposite Directions; but 2 Rest Positions would already be crushing ludicrous in this conception!). So, hypothetically, that Skinning method/algorithm calculations’ Interpolation because of the presence of additional and simultaneous Rest Position in the same 3D Rigged Model, would smooth out the Generated Artifacts from Armature Deform, preventing them to even occur (sort of), hence offering a way more realistic/organic/intuitive way of Skinning.
In other words… if we could order additional Shape Keys to become additional Rest Positions, Shape Keys would be the ultimate Rigging method. But, this would be more like a dream. I believe this ain’t so complicated; but there’s just not enough people interested (and resources involved) in making Organic Rigging real good. Out there in the CG world, so not limited to Blender only, mostly, we are still in the ‘Neolitic Age’ of CG at most, when it comes to Rigging; at least this is my opinion.
I’ve made a few experiments some time ago… which relate to this subject.
I thought that, if, instead of having a single 3D Character, I had multiple Mesh Objects for itself, but each one offering a ‘different set of a Rest Position’. So, theoretically, IF these frankenstein Mesh Objects could be regularly and smoothly swapped in realtime for 3D Animation without some sort of quaky/shaky/warping/blinking result, then these could sort of substitute the role of a lot of Corrective methods. Unfortunately (or fortunately), this experiment didn’t show for me enough potential.
When I watch and (if I) understand the video’s propositions; I might be able to think further. But for now, these is my thoughts on the subject.
PS: I’ve just watched the clip concerning “Get Pose”.
I think everytime she is mentioning a Separated or Hidden “Armature”, this is no mere Armature Object; this is a new set of Rigged Model, or Armature Object + Mesh Object (or at least an additional Mesh Object, because I do not see the need for 2 Armature Objects). If that is the case, that might have something to do with the frankenstein-like experiment I mentioned before. There would be 2 different Mesh Objects representing the Pig-Tail (one with Straight Rest Position which might be the same of the whole Body, and then an additional one for the complex Coiled Rest Position), and each one of them would have ‘their own Rig’, that is, their own Skinning and Deform Bones, but they would be integrated by some sort of Automation mechanism, like Drivers (so that as one Mesh is Masked, disappears, the other one appears seamlessly), and have their Deforms approximation smoothed out (this I’m not sure how would be achieved adequately, it was the limitations that I wasn’t able to overcome) when the Pose of one starts to match the Pose of the other. It could be I’m completely wrong (about the reality of their presented Rig) and that @Hadriscus’s idea is closer to the truth; but, whichever the case, the Rigging issues I indicated are still real.