So on a rather non-lethal note, I built a pair of robots for a predator/prey sorts of experiments[1]. I had written code that used an R/C servo to locate a beacon, and then navigate to the beacon. It worked quite well, but when I put the beacon on the 'prey' robot it rarely worked at all. The challenge was the added complexity of no fixed reference point. I found a great book[2] which described a number of approaches to the problem and ended up using a pretty simple pursuit model. The trick for me was tracking both the bearing to the target and the bearing on loss of acquisition (so keeping track of "how" it left the field of view) I ran out of CPU and memory before it got too interesting but now I should revisit it as I've got 10x the CPU and memory to play with.
[1] One thing about robots is once you get them doing the easy stuff you want them to do harder stuff.