* reset by fusion:
* state correction with tiny observation variance
* covariance matrix upate with correct observation variance
* reset wind to 0 on hard-reset during global-position-reset
increase gate
* adjust unittest: velocity gets now reset on resetGlobalPosToExternalObservation
- avoids using invalid velocity estimate to reset wind
- do not set fusion flags if starting was impossible
- reset wind to 0 before resetting velocity using airspeed if wind data
is outdated
A false positive could be triggered if velocity fusion started,
then stopped after takeoff and only position fusion started again
(because velocity fusion timed out and had a timestamp > time_last_on_ground).
We now also check if the fusion timeout is due to the innovation being
rejected (and not just a temporary check failure or data interruption).
In cold weather, fuel engines need to be warmed up. Currently, this is done by switching to stabilized mode and throttle up. The issue is that when not using GNSS data, the vehicle is not detected as "at rest" due to vibrations and cannot switch into auto/takeoff modes.
If EKF2_ENGINE_WRM is enabled, and if the vehicle is armed and landed, constant position fusion is enabled.
This allows airspeed and sideslip fusion to start during VTOL front
transition, but not in backtransition or MC. This mitigates issues seen
due to going in and out of airspeed fusion in strong headwinds in MC.
Co-authored-by: bresch <brescianimathieu@gmail.com>
* correct NED velocity according to yaw reset to keep "body-velocity" consistent
* add unit test velocityRotationOnYawReset
* '[AUTO COMMIT] update change indication'
* dont adjust velocity on yaw-reset when NE aiding active
---------
Co-authored-by: haumarco <haumarco@users.noreply.github.com>
- the estimator_aid_src::sample_timestamp would use its own timestamp to update, this
would cause the estimator_aid_src for ev velocity not to log, since the
timestamp_sample would always equal to zero and thus never be greater
than the previous published time
Co-authored-by: henrykotze <henry@autonosky.com>
* ekf2: range height skip "unhealthy" samples, but respect timeout
- we should never directly use an "unhealthy" range finder sample for
state corrections or resets, but we also shouldn't immediately abort
active rng_hgt until the timeout has passed
* check starting_conditions_passing once
* ekf2: conditional range aid change height ref
---------
Co-authored-by: Daniel Agar <daniel@agar.ca>
- if bad vertical acceleration is detected there's an emergency case
where rejected range finder observations are allowed to be used, but
this still can't happen if the sample itself is known to be bag
* ekf2: range height skip "unhealthy" samples, but respect timeout
- we should never directly use an "unhealthy" range finder sample for
state corrections or resets, but we also shouldn't immediately abort
active rng_hgt until the timeout has passed
* check starting_conditions_passing once
---------
Co-authored-by: Jacob Dahl <dahl.jakejacob@gmail.com>
* ekf: always publish baro and gnss bias even if zero
* [SQUASH] ekf2: simplify hgt bias publish
---------
Co-authored-by: Daniel Agar <daniel@agar.ca>
* ekf2: allow manual position reset when horizontal aiding is active
This allows the pilot to override position esitmates manually
* mavlink sim: add support of failure gps struck
* mavlink sim: add GNSS failure "wrong" type
* ekf2-gnss: add reset mode
This allows the user to choose whether the position should immediately
be reset to GNSS on fusion timeout or if the EKF can continue with
velocity dead-reckoning.
* ekf2: fix unit test changes due to GNSS start logic
Especially because the EKF doesn't need to reset the states if the test
ratio is already passing
* rename mode enum
* reset to gps lat lon on init
* remove obsolete reset-condition (handled in #25223)
* WIP try to upgrade compiler externally
---------
Co-authored-by: bresch <brescianimathieu@gmail.com>
Co-authored-by: Niklas Hauser <niklas@auterion.com>
bresch