Indeed, the assumption being made is basically your character is straight backed so the hips can be said to be a constant vertical distance from the head. It’s not necessarily a ‘good’ assumption because it would look pretty weird in real life but it would be interesting to see the results. In this case, the target position for the hips could be taken as being offset from the HMD by a constant amount in Z while X and Y are the same.
So this is relatively untested in a VR setting. Assuming you have an IKinema Rig set up with knee tasks and are driving the hips as you mentioned above the solver will process the data and output animations. Any results would be interesting.
As I mentioned, if you have a crouching animation you can then derive some logic that can switch your avatars animation as needed based on your human’s position. This method would give you very reliable results.