The framebyframe animation is only available as a uasset in Unreal, but if you export it from there and import into Maya on the facial rig, it won’t do much, since that animation is joint based, while you’ll need to move the facial GUI controls.
However in the UE4 animation you do see which controls are used ( as a curve animation ), so if you have a lot of patience, you can recreate the same animation in Maya triggering those controls accordingly.
Not really sure about the Pose Interpolators, never notice them.
In Maya you don’t have poses, you just have the controls that are moving the joints.
By moving those joints you create the poses, but they’re not stored or saved inside the controls, the only reason you see them as shapes inside UE4 is because of the 52 animation frame I mentioned, so each pose you crreate in Maya is “retargeted” in Unreal using the pose asset.
As far as I know the DNA file store a lot of stuff and make sure that the rig won’t break.
Regarding Maya > UE4 for the facial rig, imagine you want to have lipsync inside Unreal.
In Maya you create a set of mouth shapes that correspond to vowels/consonant ( ah, the, m, b, and so on ) which you then export to Unreal.
In Unreal, if you want to create lipsync, you can use one frame of that animation you previously created, in order to have a “ready made” mouth shape, so that you can build your own lipsync animation, rather than creating the entire mouth shape in Unreal.
The core idea behind the way the facial rig is used in Unreal is that all the hundreds of joints you see are driven by the control rig, and in order to make thing easier ( use the iPhone for example ) a simple 52 frames animation is used to create the required mouth shapes ( so basically the blendshapes from the ARKit setup ) and use that as your source for the facial animation.
Is basically using the data from the iPhone, but using them onto a different rig to achieve the same results as with blendshapes.