There are several current posts in the Sensor forum regarding compass accuracy or lack thereof. Lots of factors, including the unit's particular installation, latitude, etc, will contribute to errors. But if the repeatability is decent, anyone with patience can make their own calibration table, which should improve accuracy. Even a Stamp can do SIN & COSINE calculations (albeit with integer math). If you have a robot with wheel encoders and a decent compass calibration, what overall accuracy might be achieved when dead reckoning? That is, drive from X1, Y1 to X2, Y2,
Better yet, let's say you start from a known position & orientation called HOME. Then the robot drives an irregular pattern, either from user IR control or random wandering & object avoidance. All the while, it is keeping track of its position with trigonometry. After a few maneuvers, the robot is commanded to go directly HOME, without simply reversing all the steps accrued thus far. If the encoder units are small and the wheels don't slip much (big if), it should be able to calculate a direct vector to get it close.
Anybody experiment with this much yet? I have a nice household odometry platform (Retrobot) and this may be my next adventure, but I haven't experimented with compass sensors yet. The S2 might be a good platform too. If you have any compass experiences to relate, please chime in here!