Wandering attitude estimation in INS13GPS mode
« on: February 02, 2016, 11:38:53 am »
Hi,

I'm new[1] to librepilotbut had previously had some exposure to openpilot.

My quad flies stably in Complementary attitude estimation mode. Change to INS13GPS and the yaw, pitch, and to a lesser extent roll, sway drunkenly.

Benchtesting with green GPS & Mag produces a +-10 degree yaw variation with 15 second period, while at the same time, noise on Gyrostate.Z peaks at 0.3 deg/sec. To my mind that means at worst 7.5 * 0.3 = +-2.25 deg variation. The smoothness and constancy of the periodic variation however is disturbing, and doesn't look like integration of noise+bias. See scope pictures.

All calibrations are done. No alarms while on the bench fully powered (some mag alarms under strong throttle - the next hurdle to work through).

Re-calibrating gyros while stationary on a benchtop at 4:38:50 actually produced stronger subsequent GyroState.Z variation (so that isn't actually raw Gyro readings on the scope?). Doing a spin to make the quad to a coordinated 360deg turn didn't seem to help either (read on the forums that the EKF likes some motion?)

I later experimented with changing EKF parameters but then reverted as it only made things worse. Revo board system configuration of attitude estimation = Complementary+Mag+GPS seems to produce a stable result, but I'm unsure given that seems a hidden feature, that I should be using it.

Any and all advice and requests for more info to get stable flying in GPS mode accepted. To be honest, while it behaves like this on the bench, and doesn't look pretty taking off, I haven't gone any further really.

See attached for alarm status, scope snapshot, and UAV config export. I have the matching log, but at 2MB I just got an error in posting.

Firmware is built from Next on 2016-01-04
GIT commit hash c760276a

[1] https://forum.librepilot.org/index.php?topic=759.0

f5soh

  • *****
  • 4572
    • LibrePilot
Re: Wandering attitude estimation in INS13GPS mode
« Reply #1 on: February 02, 2016, 07:56:32 pm »
All sensors are taken in account for state estimation.
When the board is still the EKF can react with some periodic oscillations due to sensor contradictions.
Take a look here:
https://librepilot.atlassian.net/wiki/display/LPDOC/INS13+-+GPS+Navigation

If your Mag is not disturbed with motors (Green / Orange alarm) you can fly.

xfce

  • **
  • 91
Re: Wandering attitude estimation in INS13GPS mode
« Reply #2 on: March 13, 2016, 03:54:48 pm »
Hi,

I'm new[1] to librepilotbut had previously had some exposure to openpilot.

My quad flies stably in Complementary attitude estimation mode. Change to INS13GPS and the yaw, pitch, and to a lesser extent roll, sway drunkenly.

Benchtesting with green GPS & Mag produces a +-10 degree yaw variation with 15 second period, while at the same time, noise on Gyrostate.Z peaks at 0.3 deg/sec. To my mind that means at worst 7.5 * 0.3 = +-2.25 deg variation. The smoothness and constancy of the periodic variation however is disturbing, and doesn't look like integration of noise+bias. See scope pictures.

All calibrations are done. No alarms while on the bench fully powered (some mag alarms under strong throttle - the next hurdle to work through).

Re-calibrating gyros while stationary on a benchtop at 4:38:50 actually produced stronger subsequent GyroState.Z variation (so that isn't actually raw Gyro readings on the scope?). Doing a spin to make the quad to a coordinated 360deg turn didn't seem to help either (read on the forums that the EKF likes some motion?)

I later experimented with changing EKF parameters but then reverted as it only made things worse. Revo board system configuration of attitude estimation = Complementary+Mag+GPS seems to produce a stable result, but I'm unsure given that seems a hidden feature, that I should be using it.

Any and all advice and requests for more info to get stable flying in GPS mode accepted. To be honest, while it behaves like this on the bench, and doesn't look pretty taking off, I haven't gone any further really.

See attached for alarm status, scope snapshot, and UAV config export. I have the matching log, but at 2MB I just got an error in posting.

Firmware is built from Next on 2016-01-04
GIT commit hash c760276a

[1] https://forum.librepilot.org/index.php?topic=759.0

I import your uav,found you use NMEA dataprotocol, but I rember that OP/LP do not recommend this for gps flight mode.