BMP-085). I was just using some very simple low-pass filters, and results were pretty ok already:
HMC6352) also works pretty ok. It delivers very precise heading information, even when the motors are running. But, when the sensor is tilted, there is an error in heading (due to the principle of measuring). There are much more expensive sensors available that can correct for tilt, but that is not really a problem I think: In position hold, the tilt angle will always be pretty low. When doing waypoint navigation, the heading can be read from GPS, hence heading data can be calculated from complementary filtering of gyro/ GPS/ compass if necessary. Here is a extremely boring video of my DLX with pressure sensor and compass active. From second 20 till the end of the video, I am not touching throttle or yaw anymore:
Finally, I had the opportunity to test a uBlox LEA4 GPS receiver from a friend of mine. On the first picture, I was standing still for ~1 minute, on the second picture I was walking around in several "8's". This GPS receiver seems to be pretty awesome, that's why I instantly ordered one. I am pretty confident that position-hold will work out with this sensor.