search

PX4 / PX4-Autopilot

PX4 Autopilot Software
https://px4.io
BSD 3-Clause "New" or "Revised" License
8.51k stars 13.51k forks source link

way to feed in vision attitude estimate in attitude_estimator_q or LPE? #4926

Closed tuofeichen closed 7 years ago

tuofeichen commented 8 years ago

Hey guys, I'm wondering if there's a way to feed in a visual SLAM estimate of the attitude as a measurement (not just the heading, but potentially the whole rotation matrix), to attitude_estimator_q or LPE?

LPE's vision.cpp seems to only take care of translation, and attitude_estimator_q seems to only take the heading into account.

Thank you.

-FC

mhkabir commented 8 years ago

You can modify the Q estimator for that.

jgoppert commented 8 years ago

I'm gearing up to do this as well. I"ll take a look, shouldn't be too hard.

ziqi-zhang commented 7 years ago

@mhkabir @jgoppert Hello! I wonder can I feed vision estimation now? If still can't can you guys give me some advice how to modify Q estimator? Thank you very much!!!

jgoppert commented 7 years ago

Now supported.

tuofeichen commented 7 years ago

Hi @jgoppert , I was wondering where is this supported? is it in the master or any specific branch? I assume its some modifications on the attitude-q estimator, but it (master) still looks like it only takes in external heading instead of the entire rotation from vision.

Best FC

jgoppert commented 7 years ago

@tuofeichen It just takes vision yaw, that is correct, I'm working on full vision attitude support with the IEKF estimator, which should be out soon.

tuofeichen commented 7 years ago

got it, any other difference between IEKF and LPE? I'm looking at the code now but would appreciate a high level motivation of why we switch

jgoppert commented 7 years ago

IEKF is still in development and I wouldn't recommend switching yet.

Instead of separate attitude (attitude_estimator_q) and position (LPE) kalman filters, it is combined (IEKF), so it has 21 states. It also has a rough dynamic model that is found through system identification of log data. So IEKF is much more complicated but should give better results when all of the bugs are eliminated.

tuofeichen commented 7 years ago

Interesting. Got it thanks!

On Feb 9, 2017 11:46 AM, "James Goppert" notifications@github.com wrote:

IEKF is still in development and I wouldn't recommend switching yet.

Instead of separate attitude (attitude_estimator_q) and position (LPE) kalman filters, it is combined (IEKF), so it has 21 states. It also has a rough dynamic model that is found through system identification of log data. So IEKF is much more complicated but should give better results when all of the bugs are eliminated.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/PX4/Firmware/issues/4926#issuecomment-278717384, or mute the thread https://github.com/notifications/unsubscribe-auth/AGM54oLYvhbkBHqqPjJ_fgZCPPYrw7tZks5ra1DZgaJpZM4I_kgM .