For everyday users, I think they're OK. We were using Ardupilot and had very specific problems with the state estimation algos (kalman filters) for guidance, navigation, and control.
As part of our landing r&d, we tried RTK GNSS, which was generally OK in rural environments but a complete nightmare in built up urban environments. When the RTK GNSS receiver would lose fix, the system would fall back to standard GPS, thus causing a step change in position/state estimate of the drone, causing it to misbehave in flight. Fighting the "features" of the wrong tech is a slog, but I was not CTO... ce la vie.
Ultimately, we wound up licensing some tech for vision based landing.
That's fascinating, thanks! I knew navigation and control systems would be complex, but when I tried writing my own transfer functions, I got a taste for just how complex it really is. I admire what the USAF was able to do in the pre-integrated circuit era!
I guess I'm not surprised GNSS is so tricky for the final approach. Wide open spaces work great with averaging GPS, but as soon as some bogus data comes in shit goes sideways fast. I just assumed the experts figured that out, but it sounds like not the case if vision-based is the solution at that point.
As part of our landing r&d, we tried RTK GNSS, which was generally OK in rural environments but a complete nightmare in built up urban environments. When the RTK GNSS receiver would lose fix, the system would fall back to standard GPS, thus causing a step change in position/state estimate of the drone, causing it to misbehave in flight. Fighting the "features" of the wrong tech is a slog, but I was not CTO... ce la vie.
Ultimately, we wound up licensing some tech for vision based landing.