Background
I’ve noticed that with certain sensor calibration parameters, the EKF2 just fails to initialize & provide attitude estimates.
It would either:
- Not initialize at all, and there would be no local pose data (velocity, position, attitude)
- Initialize, but have erratic attitude and pose data (diverging position, completely wrong attitude)
Example log: https://logs.px4.io/plot_app?log=003e2dc9-7795-4c9a-8fc2-803f35b709c1
Sensor Calibration Parameters Reset
This issue disappears indeed when the sensor calibration data is reset (as reported in a post linked below as well).
When I did the reset, it also then allowed EKF2 to initialize correctly, and when I do the diff of parameters, it comes out like this (Left is before reset, Right is after reset):
File comparison : before_reset.txt | after_reset.txt
================================================================================
--------------------------------------------------------------------------------
Component 1-1:
--------------------------------------------------------------------------------
CAL_ACC0_XOFF : -0.314 | 0.00975
CAL_ACC0_XSCALE : 1.06 | 1
CAL_ACC0_YOFF : -0.614 | -0.00372
CAL_ACC0_YSCALE : 1.01 | 1
CAL_ACC0_ZOFF : 2.63 | -0.018
CAL_ACC0_ZSCALE : 1.47 | 1
CAL_ACC1_XOFF : -0.0503 | -0.000126
CAL_ACC1_XSCALE : 1.08 | 1
CAL_ACC1_YOFF : -0.672 | 0.00871
CAL_ACC1_YSCALE : 1.03 | 1
CAL_ACC1_ZOFF : 2.36 | -0.121
CAL_ACC1_ZSCALE : 1.51 | 1
CAL_GYRO0_XOFF : 0 | -0.00197
CAL_GYRO0_YOFF : 0 | -0.00372
CAL_GYRO0_ZOFF : 0 | -0.00245
CAL_GYRO1_XOFF : 0.00332 | 0.00269
CAL_GYRO1_ZOFF : 0.0232 | 0.0157
CAL_GYRO2_ID : 1310988 | 0
CAL_GYRO2_PRIO : 50 > X
CAL_GYRO2_ROT : -1 > X
CAL_GYRO2_XOFF : 0.000103 > X
CAL_GYRO2_YOFF : 0.000054 > X
CAL_GYRO2_ZOFF : 0.000063 > X
CAL_MAG0_ID : 0 | 396321
CAL_MAG0_PRIO : -1 | 50
CAL_MAG0_XOFF : 0 | 0.028
CAL_MAG0_YOFF : 0 | -0.0573
CAL_MAG0_ZOFF : 0 | -0.0387
CAL_MAG_COMP_TYP : X < 0
CAL_MAG_SIDES : X < 63
COM_FLIGHT_UUID : 8 | 9
LND_FLIGHT_T_LO : 154845050 | 174930794
SENS_BOARD_X_OFF : 0 | -1.18
SENS_BOARD_Y_OFF : 0 | -0.124
SENS_MAG_AUTOCAL : X < 1
SENS_MAG_AUTOROT : X < 1
SENS_MAG_SIDES : X < 63
SYS_HAS_MAG : 0 | 1
================================================================================
38 parameters differ
One of the big differences are definitely the accelerometer offset values, which seemed to have messed up the EKF2.
Question
I am wondering how tolerant EKF2 is to such bad calibration data (since even before doing the reset, it seemed like sensor_accel data contained sane output, around 9.8 m/s^2, as in the log linked above), regarding it’s initialization & pose estimation.
Would it make sense to have some kind of alert mechanism to notify user in case the calibration seems off, if that’s detectable? @bresch