That Boy was released many years before metahuman systems were created, although I think it was rigged by the same people, but the complexity of the rig doesn’t seem to by that high.
I really think they implied a universal approach for any rigs with blendshapes / morph targets, so you could even animate abstract piece of wood with beans as eyes or whatever you want - without anything resembling an actual metahuman rig. This is my assumption at this moment.
I suppose the biggest difference in Metahuman Animator compared to ARKit is that it doesn’t rely at all on iPhone’s own detection of facial parameters - it only captures raw feed (RGB + depth) and streams it to PC and then they do the tracking with their own solution (that seems to handle lips much better and be more offline quality oriented for pre-recording performances instead of real-time), so that it can also work with industry standard dual camera rigs, not just iPhones.