The same way you make them interact without AR, since AR doesn't change anything with the program except synchronize the viewpoint of the virtual camera with the viewpoint of the real world camera and give you access to a few marker locations.

Of course that's the theory. In practice, it's gonna be a tough nut to crack, especially with all the camera transformations. Moving correctly from one marker's location to the next in AR-space is a first step. I find opengl camera matrices amongst the most challenging aspects in coding so far, but maybe it's just me.

So in short, it's possible, but making 3D models interact inside an augmented reality setup, isn't exactly a suitable beginner challenge so to speak. However if you achieve something, the results can be quite cool.

For example see the work of Frantz Lasorne, such as this video.

Hello!