The nalgebra
and approx
crates have been updated to their latest versions.
- Change the default covariance update method to use the Joseph form. This is
more numerically robust at the cost of being more computationally expensive.
To use the old covariance update method, call
KalmanFilterNoControl::step_with_options()
with argumentcovariance_update_method
set toCovarianceUpdateMethod::OptimalKalmanForcedSymmetric
. - Fixed the spelling of
CovarianceUpdateMethod
(which was previously misspelled).
- Implemented tests to run basic sanity checks for the Kalman filter (with different covariance update methods), Kalman smoother, and the Kalman smoother with missing observations.
- Change primary names of many functions to match conventional Kalman filter
documentations (i.e. F for state transition model, H for observation model).
Mostly old names are still kept but are deprecated. For implementations of the
TransitionModelLinearNoControl
trait, changetransition_model
toF
,transition_model_transpose
toFT
, andtransition_noise_covariance
toQ
. The trait nameObservationModelLinear
was changed toObservationModel
and implementations of this trait need to changeobservation_matrix
toH
,observation_matrix_transpose
toHT
,observation_noise_covariance
toR
, andevaluate
topredict_observation
.
- examples do not raise clippy warnings (#6)
- Depend on nalgebra 0.26, which added initial support for const generics.