>All examples I have found assume there is a magical
>sensor that changes the output of the plant to the
>same units as the input.

Welcome to the real world :)))

Everyone seems to assume this, and that is one of the problems to solve :)

I suspect you could start by having a light weight cable to your AUV, such
as 3 cores of a ribbon cable, entering through a hole sealed with a suitable
silicon sealant. Then have some electronics inside the vehicle to control
motors and read sensors, getting control information and sending data by
high speed serial over the wires. 115k baud should do fine, and this will
allow you to measure actual sensor outputs and drive motor sensitivities
without having a lot of guess work as you go. This will allow you to use
suitable tools on a PC to collect the data and do modelling things with it
which will greatly enhance your chances of getting the autonomous operation
code working in a timely manner. Trying to sort out why your vehicle
suddenly tries to be an underwater launched rocket, or deep diving concrete
footed device is real hard otherwise, without being able to collect the data
real time. Is it sensor failure, positive feedback, strange phase shift
around the control loop, etc.

Once you get near the end of the project you make the vehicle totally
autonomous and do away with the cable and seal the hole permanently.

--
http://www.piclist.com hint: The PICList is archived three different
ways.  See http://www.piclist.com/#archives for details.