AC3.2-rc2, Flyaway due to Slow GPS Glitch Detection

I had a fairly serious ‘fly-away’ today with AC3.2-rc2 and a stock 3DR IRIS with tarot gimbal attached. Settings and configuration are the defaults for ‘IRIS+tarot gimbal’ and everything was calibrated from scratch after upgrading to AC3.2-rc2.

The flight was autonomous and everything was smooth and fine up to 3:19:59.00 PM. At that point the IRIS started aggressively flying away. Once it became clear the drone wasn’t going to recover I switched to stabilize and was able to land with only moderate damage.

Looking through the logs and checking GPS coverage it’s clear that the number of visible satellites (NSats) and GPS positioning accuracy (HDop) sharply deteriorated around the time of the fly-away. It took the GPS Glitch Detection algorithm just over 5s to spot the problem (detected at 3:20:04.40 PM), at which point the drone had already picked up quite a bit of speed.

Is this problem specific in any way to AC3.2? Is there anything that can be done to avoid this type of fly-away happening again?

I’ve read the GPS Failsafe wiki and it sounds like one way to mitigate the problem is to check for poor GPS coverage. Is there additionally any way to further improve the GPS Glitch Detection?
copter.ardupilot.com/wiki/gps-failsafe/

Thank you for any advise or tips on this!

[attachment=0]AC3.2-rc2, Flyaway due to Slow GPS Glitch Detection.png[/attachment]

I think you answered it yourself with the GPS prediction image, Andropilot the Android app from Kevin Hester also gives a warning if GPS data is likely to be dodgy before you lift.

Thank you for that. I guess Andropilot only warns you about poor GPS data (NSats + HDop) at the mission start time; as opposed to predicting problems after the mission has started?

Faster detection of GPS glitches and improved mitigation would definitely be worth exploring for future AC releases. There may also be ways to have the GPS hardware itself (uBlox, LEA-6H) report loss of accuracy more quickly. The logs show a few seconds delay between the drop in position accuracy and an increase in reported HDop.

Thanks again for the quick response!