I'm not sure how exactly it's done professionally, or even if it's done professionally, but if I were going about it, I'd use some sort of triangulation via radio receivers. Some classmates of mine did their senior design project in college on a similar
system. They used multiple radio receivers to triangulate the location of a beacon. They wrote software that would interpret the receiver strengths and guide a robot to the beacon. I wouldn't be all that hard to do the same with followspots. Now before I get flamed, I'm not saying this is easy in practice. Conceptually, it's simple enough. In practice, the software is incredibly difficult in terms of filtering out noise and selecting the "clean" signal. As with any triangulation
system, the accuracy improves with the more receivers you have, but at some
point increasing the number of receivers becomes impracticle. Also, the more receivers you have, the more processing
power is required to sift through all the data. If you can come up with a "clean" signal, converting that to
DMX instructions for a moving head is probably one of the easier parts.
In my opinion, the hardest part of this concept is the data filtering algorithms. You've got to filter and determine a "clean" signal, at what, 40 times a second, to keep up with the
DMX refresh rate. Honestly, I think this would make for a great electrical engineering design project (I wish I would have thought of it when I did mine), but in practice, it seems it isnt really used.
Someone please correct me on any of this if I'm off-base.