Hello Shuo,
Regarding the example training model:
The follower frame is rotated so that we can align the base and follower frames. For a planar joint, X and Y axes are ALWAYS prismatic and Z axis is revolute. So in the example you can check that the base frame of the planar joint has X pointing up, Y pointing to the right and Z pointing into the screen.
For the planar joint to work the follower needs to have the same configuration to make sense. However the 'Rim' coordinate frame has Z pointing to left and X,Y pointing out of the screen and up respectively.
We need to make this second frame look like the first one. First of all translate this to the tire contact point along the Y axis (by half the diameter of the tire), then for the planar joint to work, we need to rotate it so X is pointing up, Y is pointing to the right and Z is pointing out of the screen. We use the aligned axis method to do this rotation.
Your model:
Start small. Your model is already too big to try to debug. Add assemblies one by one. For e.g. add only the front left assembly first. Don't add everything at once. When you have one working then add the next one to see what happens. When you model this way incrementally you can catch errors better since it is easier to debug.