An autopilot use heading information from a fluxgate compass so of course that will be used as a feed-back. However in my understanding of an autopilot it usually consists of two cascaded loops. The first one for heading and the other one for setting the rudder position. The first one use heading as feed-back and the last one a "rudder feed-back unit". Some autopilot manufacturer has taken away the need fro that last unit and I just wander how that can be done.
Now I understand your question. Yes cascade control where the rubber has it's own PID control function allows one to optimize the controller tuning settings for the rudder controller and the heading controller independently such that the heading controller outputs a 'desired setpoint' to the rubber controller as it's PV measurement.
In such a configuration the rubber controller is said to be the slave controller and the heading controller the master controller. It gives better controllability. However with more the powerful control possibilities with a microprocessor this can all be done inside the micro and could take advantage of feed-forward or adaptive or other 'smart' control algorithms. So I guess it's a trade off, saving the cost of the hardware needed to generate a rudder position measurement while still having good control behavior by compensating by using more advance software control features.
As a hardware guy, I would prefer having actual rubber feedback measurement in that it then is easy to add additional alarm functions if say the rubber linkage was to break and the actual rubber commands were not being performed.
Lefty