I’ve been out of quads for a few years. I had a near disaster with a large quad that went rogue, it went about 1/2 a mile off course, but it somehow miraculously came back and crash landed near the launch point. I suspect something with the internal magnetometer may have been part of the cause. FC was a HK32 pixhawk knockoff and I often got compass calibration errors. But sometimes it would work perfectly fine on long mission paths (maybe a mile round trip).
But I’ve been thinking about how to have a good backup for RTL. Could you have the quad take pictures, say pointed down. And then if GPS/compass/radio fails, it could try to back track via the picture history back to launch?
I know photogrammetry takes a lot of computing power, so it would have to be some algorithm that a small Raspberry PI could handle.
I was also wondering if a system could be built that had preloaded Google earth images to allow navigation from just the loaded pictures and an onboard camera. Obviously sun/clouds/seasons/time of day could screw all that up.
I’ve been trying to catch up on all the latest. It seems some Non-GPS navigation is done by just counting up all the accelerations/speeds to figure out where it is, but that seems error prone to me.
Probably way over my head to develop something like that, so maybe somebody else has done it or tried it.
Yes you are not the first person to think about it, and yes there is a solution… it`s called SLAM, google for it. But you need a lot more computational power than a Rpi.
Just get a better and redundant GPS it`s cheaper and more robust.
Yes, it is called SLAM and it exists for quite some time now. Despite there are variants of the algorithm that require low computing power, I do not see it being implemented and used with ardupilot yet. First of all I do not think it is necesary in 99% of applications. Better solution for what you need is to get a better FC, dead reckoning apparently works great and it will give results depending on the quality of your FC IMU.
Again, another gps would be a good option but adding a camera that would be looking down, companion computer etc is not worth it.
I mean, I am sure that some researches studied. SLAM algorithms on ardupilot drone but i do not see it being used soon. Though it would have a nice application in e.g. indoor flying.
