* EKF: Improve covariance prediction stability
Eliminates collapse of vertical velocity state variance due to rounding errors that can occur under some operating conditions.
* EKF: Fix typo
* test: Fix initialisation test cases
Provide sufficient time for variances to stabilise and fix calculation of reference quaternion for alignment.
* test: Allow for accumulated rounding error in IMU sampling test
* test: Allow sufficient time for quaternion variances to reduce after initial alignment
* test: Increase allowance for tilt alignment delay and inertial nav errors
* test: Increase allowance for tilt alignment delay and inertial nav errors
* adpat reset velocity test
* test: update change indication file
* test: Adjust tests to handle alignment time and prediction errors
* README.md: Add documentation for change indicator test
* EKF: centralize range finder tilt check
* Ekf-control: do not double check for terrain estimate validity
isRangeAidSuitable can only return true if the terrain estimate is valid
so there is no need for an additional check
* range_finder_checks: restructure the checks to avoid early returns
There is now only one clear path that can lead to the validity being
true.
Furthermore, if the _rng_hgt_valid is true, we can trust it and we don't
need for additional checks such as tilt.
The case where we need to provide fake measurements because the drone is
on the ground and the range finder data is bad is already handled
in "controlHeightFusion" so there is no need to hack the range finder
checks with that.
* Add Sensor and SensorRangeFinder classes
The purpose is to encapsulate the checks for each sensor in a dedicated
class with the same interface
* SensorRangeFinder: encapsulate in estimator::sensor namespace
* EKF: rename _sensor_rng to _range_sensor
* Range checks: include limits in valid range
* RangeChecks: update comment in the continuity checks
* RangeChecks: move more low-level checks in functions
Also move setTilt out of the terrain estimator, this is anyway protected internally
to not compute cos/sin if the parameter did not change.
* Sensor: remove unused virtual functions
Those are not required yet but can still be added later
* SensorRangeFinder: re-organise member variables
Also rename getRangeToEarth to getCosTilt
* SensorRangeFinder: split setSensorTilt and setCosMaxTilt functions
* SensorRangeFinder: Add a few unit tests
- good data
- tilt exceeded
- max range exceeded
* SensorRangeFinder: set hysteresis in us instead of ms
* SensorRangeFinder: Add more tests
* SensorRangeFinder: update continuity, hysteresis and stuck tests
* SensorRangeFinder: rename variables
* SensorRangeFinder: get rid of "delayed" specification
From the SensorRangeFinder class point of view, it's not relevant to
know if the data is delayed or not
* SensorRangeFinder: move time_last_valid out of stuck check
* SensorRangeFinder: rename file names to sensor_range_finder
* SensorRangeFinder: address Kamil's comments
* SensorRangeFinder: Add more tilt tests
* SensorRangeFinder: store current tilt offset
This is to avoid recomputing cos/sin functions at each loop
* EKF: Enable GPS flight without magnetometer
Enables takeoff in a non-GPS flight mode with mag fusion type set to MAG_FUSE_TYPE_NONE. After sufficient movement the EKF will reset the yaw tot he EKF-GSF estimate. After that GPS fusion will commence.
* EKF: Fix unconstrained yaw and yaw variance growth when on ground
* EKF: Ensure first yaw alignment can't be blocked
* EKF: Increase yaw variance limit allowed for alignment
Flight test data indicates that an uncertainty value of 15 deg still provides a reliable yaw estimate and enables an earlier alignment/reset if required.
* EKF: Remove unexecutable code
* EKF: Restructure heading fusion
* EKF: parameterise quarter variance check and retune default value
* EKF: Pass by reference instead of pointer
* EKF: Clarify reset logic
* EKF: Remove incorrect setting of mag field alignment flag
* EKF: Non-functional tidy up
* EKF: Fix non-use of function argument
The updateQuaternion function was using the _heading_innovation class variable instead of setting it using the innovation argument.
* EKF: Fix undefined variable
* EKF: Use single precision atan2
* EKF: remove unnecessary timer reset and unwanted execution of reset function
* EKF: Don't declare a mag fault when non-use is user selected
Doing so produces unnecessary user alerts.
Replace the multiple if-else statements by a generic equation.
- For a multicopter, the attenuation factor is 2 and symmetrical
- For a fixedwing, the attenuation factor is 1 if the acceleration is
positive and that centripetal correction is available and 2 otherwise.
Note that the function "sq" needs to be const in order to be used in a
const function.
* ekf2: centralize mag covariance reset. A complete and clean reset of the
mag states covariances is now performed through the "resetMagCov"
function only. This avoid having slight differences of implementations
across the code.
* Ekf: reset quat cov with initial uncertainty instead of zero
* Add const modifiers
* Pass imu data as sampe
* Remove emergency reset request counter
* Improve matrix library usage
* Indentation fix
* Do not store innovation matrix for each model
* Make weights a separate vector
* EKF: Add missing alignment transfer to AHRS solutions
* EKF: Replace #define constants with static constexpr
* EKF: Move declaration for weights into GSF section
* EKF: Fix documentation error
Co-authored-by: Paul Riseborough <p_riseborough@live.com.au>
* EKF: Use common rate vector calculation for offset corrections
* EKF: Remove duplicate matrix entry calculations
* EKF: Create a EKF-GSF yaw estimator class
* EKF: add emergency yaw reset functionality
* EKF: remove un-used function
* EKF: Ensure required constants are defined for all builds
* EKF: Fix CI build error
* Revert "EKF: remove un-used function"
This reverts commit 93005309c7f3794414ad99c86218b3062e00bbd3.
* EKF: Replace in-lined Tait-Bryan 312 conversions with function call
Also remove unnecessary operations
* EKF: Remove unnecessary update of external vision rotation matrix
* EKF: Use const
* EKF: use const
* EKF: don't use class variable as a temporary variable
* EKF: update comments
* EKF: Improve efficiency of yaw reset
Use conversion from rotation matrix to Euler angles instead of quaternion to euler angles.
* EKF: use const
* EKF: remove un-used struct element
* EKF: more descriptive function name
* EKF: use existing matrix row operator
* EKF: remove unnecessary rotation matrix update
* EKF: Use square matrix type
* EKF: Improve protection for bad innovation covariance
* EKF: Use matrix library operations
* EKF: Replace memcpy with better alternative
memcpy bypasses compiler sanity checks and is unnecessary in this instance.
* EKF: Split EKF-GSF yaw reset function
Adds a common function to support yaw reset that can be used elsewhere.
* EKF: Use common function for quaternion state and covariance yaw reset
* EKF: Replace inlined matrix operation
* EKF: Use const
* EKF: Change accessor function name
* EKF: Use const
* EKF: Don't create unnecessary duplicate variable locations
* EKF: Remove duplicate covariance innovation inverse
* EKF: Don't create unnecessary duplicate variable locations
* EKF: Rely on geo library to provide gravity
* EKF: Improve protection from bad updates
* EKF: Reduce effect of vibration on yaw estimator AHRS
* EKF: Improve yaw estimator AHRS accuracy during manoeuvre transients
* EKF: do not fuse multiple times the same height
The _fuse_height flag was never set to zero, hence the fusion was called
at each iteration, even if no new data is available.
The effects were: high CPU usage and virtually less measurement noise
due to multiple fusion of the same sample
Also remve unused variables
In a velocity reset we only used the GPS measurement if _gps_check_fail_status.value was equal to zero. The value of this flag is independent of EKF2_GPS_CHECK so checks can fail even if they are not configured to have any effect.
To obtain the correct difference between two angles, we need to wrap
the result between -pi and pi. Otherwise, the difference between two
angles close to 180 degrees but with opposite signs will produce a large
error.
For example if a = -179 and b = 179, b - a = 258 instead of -2 degrees
As the name can be ambiguous, it gets renamed
"uncorrelateQuatFromOtherStates".
Also replace the loops storing the values and reapplying them by simply
zeroing two slices (this also saves 130B of flash).
The intent of the function "uncorrelateQuatStates()" is to uncorrelathe quaternions from the other states but not between each other as they are and should remain linked alltogether.
Clearing the quaternions to quaternion covariances introduced unstabilities in other states (especially accel biases).
bresch
Paul Riseborough
kamilritz
Matthias Grob
Daniel Agar
TSC21
CarlOlsson
Paul Riseborough