ianwood said:So, it seems we are at about 80% of a solution with some caveats. Higher declination areas are still seeing more than others. Sar104's results still need to be factored in but they seem to be much lower than 80%. I need to retest course lock. The one time I tried it, it worked pretty much fine. And I'm having yaw issues.
On my yaw issues, I just started editing some of my footage from the weekend. I am definitely seeing some new un-commanded yaw that's not just occurring during hover. It seems to be coming from the FC as the gimbal is not compensating for it. This is a bit of a concern as it is like having the limitation of a 2D gimbal all over again just in a new way.
I re-ran my tests again this evening with a more careful setup - specifically preparing a completely level surface for startup and also with a magnetometer survey to check for local magnetic field anomalies. I also made reasonably precise bearing measurements in both GPS and CL mode, and I think I have reasonable video, although I'll have to work on that tomorrow morning. The bottom line, however, is that I can confirm my observations from yesterday, including the uncommanded yaw at takeoff, which appears to be equal to the local declination to the nearest degree or so, the somewhat random reappearance of the J-hook after the initial learning phase cleared it up, and a persistent course lock error of twice the declination.
I also noticed the distinct lack of altitude control in forward (or backward) flight that was mentioned by a couple of others.
I'll reiterate that I do not believe that this is anything other than a patch to improve (speed up) the AP learning algorithms, and the continual recurrence of the double declination offset seems to indicate that DJI have been unable to locate the fundamental source of the problem, which is really rather mystifying. I simply cannot imagine how this can be so hard to find and fix.
I'll post some video tomorrow morning that hopefully will illustrate these effects. On the plus side I did get to play with the beta version of the Lightbridge iOS app, which seems more responsive than the Android version that I'm running on a Nexus 7.