The machinedesign.com video, showing proportional control of position for pneumatics, is very similar to what's going on inside a hobby servo. The idea is to feed back an error signal, which is the difference between the reference / set point (where you want it) and the output (actual position). The error signal is multiplied with the reference to shift the output, and this reduces the error until the position is correct.
Hobby servos have all the electronics and actuators built in, but in pneumatics, you've got to link up the valve controller (reference) with the cylinders (output)- as these are separate systems. Additionally, you have to have a sensor on the cylinder to determine the actual position for calculating error.
Ok, so say you have all of that. Now how do you generate a reference? Say you want the cylinder to extend 25%, if you try to do pulse width modulation (on for 25% of the time, off for 75% of the time), you're going to run into the devastating compressibility of air that the Mad Hatter was talking about. I found this out the hard way.
One way around this is to use a voltage to pressure transducer, like the Bellofram T-1000. Similar to a typical pneumatic valve, but it's regulator that can provide a set pressure, rather than just on & off. Now your reference is a pressure value, and your electronics need only increase the voltage (or current - different model) until the cylinder position is where you want it. I haven't seen anyone build this particular setup, but I think it would work.
Worth noting that this setup would only work for mechanisms that require more force for more extension, but that's typically the case, anyway. I've bought the T-1000s on ebay for about $80 each.
The problem with trying to use pressure to regulate position, is that normally in animatronics it is not a controllable ratio. Most of the time when there is enough pressure for they cylinder to begin moving, it will continue to do so. This could potentially work if the load was directly centered above the force of the cylinder.
You're right. It would depend on the mechanism it was controlling. For this setup, a single-acting cylinder (spring return) would probably cooperate better, too.