Apple Park Visitor Center AR App

After watching many online video of Apple Park Visitor Center AR app, I cannot figure out how can they achieve such position precious 3D animation mapping to a real life 3D model. As you can see they are using Unreal Engine, I think we can achieve the same effect using Unity 3D?

1. How do they detect the Plane? Although there is a UI (Slowly pan across the landscape to begin initialization) guiding user to pan the iPad to a surface for initializing a Plane, it looks like the user do not need to pan the iPad and Anchors are pinned. The real life 3D model even is not a flat surface! From my ARKit development experience, I cannot achieve the same result.
2. How do they map 3D animation exactly the same size of the real life 3D model? I can understand that the scale size can be predicted, how about the orientation? Core Location (iBeacon/WiFi/GPS/Compass heading)? Core ML + Vision (trained object detection)?
3. After the iPhone X was announced, there are a few new ARKit APIs. Are they using the new ARTrackable? Or simply a ARSCNViewhitTest?

I am so interested in achieving the same result as the Apple Park Visitor Center AR app. Please share your thoughts.

 

A video showing ARKit + Vision:

The above example showing how to make use of Vision in ARKit to observe and replace rectangles. But only rectangles. Can we use Vision to find out a real life non rectangluar objects?

 

How do you know they are using ARkit?

Everything I see in the two videos is achievable with a customized version of marker-based tracking (or model-based tracking), which has been around well before ARkit.

 

Thank you for your information.
I try to search more information regarding to model-based tracking AR, and I find that the Apple Park AR demo is amazingly fast and accurate. Could you suggest any framework/SDK/plug in? I really want to build something as amazing as the demo app.

 

They have a pre-trained marker of a 3D object, namely, the 3D real world to-scale replica.

Vuforia Object Targets https://library.vuforia.com/articles/Training/Object-Recognition

 

Did you find how can we achieve Model-based tracking AR using Unity and ARKit ? any help will be great to us ... thanks