Struct adskalman::KalmanFilterNoControl

pub struct KalmanFilterNoControl<'a, R, SS, OS>
where
    R: RealField,
    SS: DimName,
    OS: DimName,
{ /* private fields */ }

A Kalman filter with no control inputs, a linear process model and linear observation model

Note that the structure is cheap to create, storing only references to the state transition model and the observation model. (The system state is passed as an argument to methods like Self::step.) Given the lifetime bound of this struct, a useful strategy to avoid requiring lifetime annotations is to construct it just before Self::step and then dropping it immediately afterward.

Implementations

impl<'a, R, SS, OS> KalmanFilterNoControl<'a, R, SS, OS>

where R: RealField, SS: DimName, OS: DimName + DimMin<OS, Output = OS>, DefaultAllocator: Allocator<R, SS, SS> + Allocator<R, SS> + Allocator<R, OS, SS> + Allocator<R, SS, OS> + Allocator<R, OS, OS> + Allocator<R, OS> + Allocator<(usize, usize), OS>,

pub fn new( transition_model: &'a dyn TransitionModelLinearNoControl<R, SS>, observation_matrix: &'a dyn ObservationModel<R, SS, OS> ) -> Self

Initialize a new KalmanFilterNoControl struct.

The first parameter, transition_model, specifies the state transition model, including the function F and the process covariance Q. The second parameter, observation_matrix, specifies the observation model, including the measurement function H and the measurement covariance R.

pub fn step( &self, previous_estimate: &StateAndCovariance<R, SS>, observation: &OVector<R, OS> ) -> Result<StateAndCovariance<R, SS>, Error>

Perform Kalman prediction and update steps with default values

If any component of the observation is NaN (not a number), the observation will not be used but rather the prior will be returned as the posterior without performing the update step.

This calls the prediction step of the transition model and then, if there is a (non-nan) observation, calls the update step of the observation model using the CovarianceUpdateMethod::JosephForm covariance update method.

This is a convenience method that calls step_with_options.

pub fn step_with_options( &self, previous_estimate: &StateAndCovariance<R, SS>, observation: &OVector<R, OS>, covariance_update_method: CovarianceUpdateMethod ) -> Result<StateAndCovariance<R, SS>, Error>

Perform Kalman prediction and update steps with default values

If any component of the observation is NaN (not a number), the observation will not be used but rather the prior will be returned as the posterior without performing the update step.

This calls the prediction step of the transition model and then, if there is a (non-nan) observation, calls the update step of the observation model using the specified covariance update method.

pub fn filter_inplace( &self, initial_estimate: &StateAndCovariance<R, SS>, observations: &[OVector<R, OS>], state_estimates: &mut [StateAndCovariance<R, SS>] ) -> Result<(), Error>

Kalman filter (operates on in-place data without allocating)

Operates on entire time series (by repeatedly calling step for each observation) and returns a vector of state estimates. To be mathematically correct, the interval between observations must be the dt specified in the motion model.

If any observation has a NaN component, it is treated as missing.

pub fn filter( &self, initial_estimate: &StateAndCovariance<R, SS>, observations: &[OVector<R, OS>] ) -> Result<Vec<StateAndCovariance<R, SS>>, Error>

Kalman filter

This is a convenience function that calls filter_inplace.

pub fn smooth( &self, initial_estimate: &StateAndCovariance<R, SS>, observations: &[OVector<R, OS>] ) -> Result<Vec<StateAndCovariance<R, SS>>, Error>

Rauch-Tung-Striebel (RTS) smoother

Operates on entire time series (by calling filter then smooth_from_filtered) and returns a vector of state estimates. To be mathematically correct, the interval between observations must be the dt specified in the motion model. Operates on entire time series in one shot and returns a vector of state estimates. To be mathematically correct, the interval between observations must be the dt specified in the motion model.

If any observation has a NaN component, it is treated as missing.

pub fn smooth_from_filtered( &self, forward_results: Vec<StateAndCovariance<R, SS>> ) -> Result<Vec<StateAndCovariance<R, SS>>, Error>

Rauch-Tung-Striebel (RTS) smoother using already Kalman filtered estimates

Operates on entire time series in one shot and returns a vector of state estimates. To be mathematically correct, the interval between observations must be the dt specified in the motion model.