We're competing in the Outback Challenge and have a strange issue with our GPS setup. Turning on the three redundant APMs one always gets a GPS lock within a few seconds of starting, but the other two can sit for half an hour or more under a clear sky without acquiring a lock. The only work around I've found is to disconnect the I2C compass cable from an APM with no lock until the lock is acquired, and then reconnect. Even after this the two problem GPSs often see less than half the number of satellites the working GPS sees.
The APMs are mounted upside down, with the 3DR UBlox/Compass modules the right way up. Since the compasses are normally ROLL_180 because they're upside down in the housing but because they're the opposite orientation of the APMs we use ROLL_NONE, as per setup instructions. The compasses have all been calibrated using the compass dance. I don't understand the compass mot calibration but I assume that it doesn't matter if we're not electric powered.
The GPSs have all been upgraded to the latest settings using the u-Centre instructions on arducopter, and connected to u-Centre they all have identical performance, with 3D accuracy around 4.5 metres.
The APMs have identical parameters, and are all running the same firmware based on 3.0.3 (we commented out some modes and code we didn't need, including compass declination tables in case that's pertinent to this problem - we set the declination manually to -12 30 for Sydney Australia) because we couldn't compile and stay within upload size limit (this is another question).
My guess is that the compasses are used to init the GPSs somehow in the APM code, and some combination of the orientation of the compass to the APM or being in the Southern Hemisphere has somehow mucked this up.
You can see the tray with the GPSs and APMs setup at 4:17 in this video:
We can fly once we have the locks, but getting the locks in the first place can be complex. Any pointers or suggestions would be really appreciated.