das wurde so auch schon bei den google groups angesprochen. Keine Ahnung, warum die Devs das noch nicht angegangen haben. ....
Aus drones-discuss:
Re the GPS glitch protection, I've implemented velocity based glitch protection in my clone's gps_glitch1 branch. The important parts are:
- the velocity based glitch protection is in AP_InertialNav.cpp
- the arducopter simulator has an array of lat/lon values that are applies at 1 second intervals during the gps-glitch test. The values used in this test are taken from an actual user's log (the user was Garrick).
I've also created a little video of how it works in the simulator: http://www.youtube.com/watch?v=o0U6Zc5Pv3g. In short, you can see it works well at first but if the glitch persists for more than a few seconds the circle-of-possibility expands to accept the new glitchy position and the copter will fly off. Then as the gps position returns quickly to it's real position, the copter again rejects the gps position for a few seconds meaning that it's actually slower to recover vs the unprotected algorithm.
So it kinda works but we might need to improve the algorithm further. The idea that I'm working on now is one Leonard and I discussed yesterday which is to use a max acceleration based circle-of-possibility instead of a max velocity based one.
- the velocity based glitch protection is in AP_InertialNav.cpp
- the arducopter simulator has an array of lat/lon values that are applies at 1 second intervals during the gps-glitch test. The values used in this test are taken from an actual user's log (the user was Garrick).
I've also created a little video of how it works in the simulator: http://www.youtube.com/watch?v=o0U6Zc5Pv3g. In short, you can see it works well at first but if the glitch persists for more than a few seconds the circle-of-possibility expands to accept the new glitchy position and the copter will fly off. Then as the gps position returns quickly to it's real position, the copter again rejects the gps position for a few seconds meaning that it's actually slower to recover vs the unprotected algorithm.
So it kinda works but we might need to improve the algorithm further. The idea that I'm working on now is one Leonard and I discussed yesterday which is to use a max acceleration based circle-of-possibility instead of a max velocity based one.
Zuletzt bearbeitet: