When a state stops to be estimated it becomes a "consider state". Its value and variance are frozen but its covariance with other states continue to evolve normally.
- removes conservative accel bias variance limiting
- force symmetry is skipped after fusion of NED vel/pos (a direct measurement)
---------
Co-authored-by: bresch <brescianimathieu@gmail.com>
* ekf2_derivation: use single source of state definition
The state is defined as an ordered dictionary of group elements and
everything else is generated using that state definition
* ekf2: generated state sample add const reference getter
---------
Co-authored-by: bresch <[brescianimathieu@gmail.com](mailto:brescianimathieu@gmail.com)>
Co-authored-by: Daniel Agar <daniel@agar.ca>
- split mag_3d into new standalone mag fusion and mag fusion allowed to update all states (full mag_3d)
- new dedicated control logic for mag/mag_3d fusion and standalone mag heading fusion
- if WMM available use for mag_I and mag_B init
- mag states reset if external yaw reset (yaw estimator, GPS yaw, etc)
- mag reset if declination changed (eliminate _mag_yaw_reset_req)
- mag fusion (but not mag_hdg or mag_3d) can be active during gps_yaw or ev_yaw (if yaw aligned north)
Co-authored-by: bresch <brescianimathieu@gmail.com>
* ekf2-test: remove outdated codegen comparison
The definition of states changed so the comparison with the old
derivation cannot work anymore.
---------
Co-authored-by: bresch <brescianimathieu@gmail.com>
There is no reason to contrsain the position uncertainty, estpecially
when flying with velocitiy aiding only.
Note that all the variances are already contrained to sane values at the
end of the covariance prediction
- pass new airspeed sample around when available
- can't completely eliminate _airspeed_sample_delayed until resetWind()
called from sideslip fusion is updated
- refactor all EKF backend output predictor pieces into new OutputPredictor class
- output states are now calculated immediately with new high rate IMU rather than after EKF update
- IMU delayed sample is passed as around as control data to avoid storing an extra copy and make the requirement clear
- a growing number of samples come into the backend with the time
already delayed (sensor's interrupt setting timestamp sample)
- if the incoming timestamp is already delayed then the new data checks
(relative to latest IMU) can be slightly wrong
- handle almost all timestamps and checks on delayed time horizon,
except for explicit checks of new samples
- isRecent() and isTimedOut() helpers use delayed time
- add new isNewestSampleRecent() used for checking the incoming
timestamp of the incoming (adjusted) data
The noise spectral density, NSD, (square root of power spectral density) is a
continuous-time parameter that makes the tuning independent from the EKF
prediction rate.
NSD corresponds to the rate at which the state uncertainty increases
when no measurements are fused into the filter.
Given that the current prediction rate of EKF2 is 100Hz, the
same tuning is obtained by dividing the std_dev legacy parameter by 10:
nsd = sqrt(std_dev^2 / 100Hz)
bresch