When magnetic field states have been reset in-flight using a single sample, the magnetic field states are not used to constrain heading drift for a period after the reset. This period has been shortened from 10 to 5 seconds which is enough time to average out the effects of measurement noise (the original concern). The shorter time has enabled the previous practice for RW vehicles of using magnetic heading in that time period to constrain yaw drift to be discontinued. This is necessary becasue while magnetic heading is being used, it fights the yaw corrections obtained from GPs observations and lengthens the time required to recover from a bad mag calibration.
If state errors were large before the reset, then failure to reset the covariance matrix terms can result in incorrect fusion of position and velocity measurements after the reset due to inconsistencies in the covariance matrix.
When performing the initial in-flight mag yaw reset for RW vehicle, do not reset the quaternion states and corresponding variances unless there has been a change in yaw angle large enough to cause problems with navigation.
This is because the state estimates after a reset are more vulnerable to transient sensor errors, so a reset should be avoided if possible.
When performing the initial in-flight magnetic field reset for fixed wing vehicles, resetting the quaternion states and their corresponding covariances should be avoided unless yaw errors are large, because state estimates are vulnerable to transient sensor errors immediately following a reset.
Ensures that a complete reset of velocity and position states will always be performed if yaw has had to be reset using GPS velocity.
Ensures that the yaw_align status cannot be set to false once the filter has aligned.
Reduces susceptibility to incorrect estimation of acceleration bias during sustained yaw rate.
Requires an increase in RAM allocation of 837 Bytes to allow for the longer IMU and output predictor buffers that can be created.
Use of air data to navigate should be classified as dead-reckoning because neither ground relative velocity or position is observed directly and errors grow faster.
Daniel Agar
Paul Riseborough
ChristophTobler
CarlOlsson