EDIT: new version of the arm: https://www.youtube.com/watch?v=wQ6FGeSjN9c
I've built a robot arm (using OWI 535, 4 deg. of freedom) that uses a hierarchy of control loops for positioning - there is no inverse kinematics. There are three levels of control. The first level controls the speed of the motors using potentiometers as sensors. The second levels loops control the angles of each joint. Loops on the third level coordinate second level loops in complex movements such as extending reach while maintaining hand pitch.
It's based on PCT (perceptual control theory) and a fascinating virtual model of a human arm (http://www.billpct.org/ after dl, look for Coordination Demo)
I'ts primary purpose is modeling the human nervous system, and since there is no inverse kinematics, just simple control loops, it might be useful elsewhere. The control loops are currently simulated on the computer. An arduino collects sensor data and sends it to the computer, and receives motor speed commands and sends them to motors. It might also be possible to do all the calculations on the arduino itself (but then I'd need more analog inputs). The software is written in C#.
So.. it's working. I'm planing a more complex arm. I'd be happy to answer any questions.