I'm running 2.6 Delta on a pretty standard tricopter. In stabilize mode I get a mostly rock solid hover. However, every 20 to 30 seconds, even when I'm just hovering 6ft AGL, there's a pretty large hiccup or twitch where the rear motor slows significantly. Sometimes it just causes the craft to dip backwards and recovers, but sometimes it is so severe that it falls out of the air.
I had the issue with 2.5.4 and 2.5.5 as well. I've replaced the ESC and switched the motor with one of the other arms, and re-checked all wiring, and it still happens on the rear motor. This has happened with two different APM2 boards even, so I'm pretty sure it's not hardware at this point.
I started trying to adjust the PID values and thought at first it was helping. First thing I did was set rate_D on pitch to 0 since I've read D has contributed to twitchiness in the past. I also started backing down rate_P, from 0.180 all the way to 0.9. I also switched to acro mode to narrow the problem down further, and it was still happening.
It may be my imagination, but it seemed that the last test, with lowered rate_P for both pitch and roll, the problem was getting worse the longer I'd hover, and the problem seemed to start happening in roll but not quite as bad as the pitch.
Any thoughts on what to look at next? I was watching the tuning graph and the RC input looks fine, so it's not a radio issue. Also the sensors all look fine, so it doesn't look like an attempt to compensate for a whacky sensor or vibration issue. The motor output for ch4 looks like it does drop out, but is that the cause or a reaction/symptom? I don't see any correllating change in input or sensor.
The fact that it'll hover rock solid for 20 to 30 seconds (and sometimes even for a minute or more) confuses me, which is why I originally looked for intermittant hardware or wiring failures. I've even done some pretty aggressive flying and often it flies just fine (though I've seen it happen during aggressive flying too).
My limited understanding of the PID math made me think that this was some error build up that reaches a threshold and over-corrects, but I'm not sure where in the numbers this would happen, and I think I've eliminated most of that possibility (i've had rate_P low and rate_I and rate_D zero).
This is pretty frustrating, since I've already had to send my board back twice for the diode issue and a subsequent uart failure, and now that I've finally got a functional board, 2.6 Delta is otherwise amazing. I'm all decked out with FPV gear, but I can't risk flying more than a couple feet off the ground because it WILL eventually tank.
Any suggestions? I'll try to catch a tlog tomorrow, in case that can help.