With regard to the phenomenon known as "Fly-away", where a multi-rotor craft seems to just take off on a strange course and crash or get lost, I think the most likely culprit is the GPS. I have seen "fly-away" happen on the bench, using an APM 2.5.2 controller and LEA-6H GPS module. With good GPS signal and 3D position, I saw the real-time indicator in Mission Planner begin moving wildly away from my location. Since it was really just sitting on my bench, I didn't panic and just watched it for a while, and it "flew" nearly a mile away before I got tired of watching. I suspect what I watched was the flight controller reading bad input from the GPS, and had the equipment been airborne, my quad would have flown quite a distance in an attempt to correct the false positioning readings I watched. I believe that flight controllers place too much weight on GPS data, even to the point of overriding all other sensors, even when those sensors all agree the craft is not moving. The programming of flight controllers that use GPS should be refined such that conflicting instrument readings trigger a pilot confirmation sequence. When the deviation of instrument readings exceeds a preset threshold, I believe the controller should either send a signal back by a telemetry path, or turn on a strobe to alert the pilot that an error condition has been encountered. The pilot could then move a three-position switch down (from center) to indicate that the craft was in fact in a relatively stable hover, and the controller should attempt to resume normal flight, or the pilot could choose the up position to indicate the craft is indeed in "fly-away", and the flight controller should discontinue all stabilization efforts and let the pilot fly it to a safe landing. The strobe (if used) could be located where it is visible from an FPV camera as well as the ground. Does anyone else have any insight into the fly-away problem?