• Kalman Filter
  • In Categories Of
  • Algorithm
  • Robotics
  • Tagged With
  • Robotics
  • Signal Processing
  • ##Purpose and Usage

    • Eliminate noise in measurements
    • Generate non-observable states(e.g., Velocity from position signals)
    • For prediction of future state
    • Optimal filtering

    ##Framework and Model

    ###Given:

    • A discrete stochastic linear controlled dynamical system:

    Each current signal value x^k is a combination of previous signal value $x_{k-1}$ times a constant, a control signal $u_{k}$ and a process noise and a process noise signal $w_{k-1}$(which usually considered as zero).

    • A measurement function, where $v_{k}$ is the measurement noise.

    • Assume the process noise and the measurement noise are both considered to be normal distribution that

    In reality, covariance matrix Q and R may change in every iteration. We assume they are constant here however.

    ###Goal:

    Find the best (recursive) estimate of the state x of the system.

    ###Computational Origins

    Define $e_{k}^{-}$ to be a priori state estimate at step k given knowledge of the process prior to step $k$, and define $e_{k}$ to be a posteriori state estimate at step $k$ given measurement $z_{k}$. Then a priori and a posteriori estimate errors can be defined as:

    The a priori estimate error covariance is then

    The a posteriori estimate error covariance is then

    Then How can we optimally (linearly) combine the estimate and measurement to obtain the best reconstruction of the true x? The answer given in “The Probabilistic Origins of the Filter” found.

    Where residual is $z_k - H \hat{x}_{k}^{-}$. It also can be called as measurement innovation.

    The Kalman filter gains are derived by minimizing the posterior error covariance, resulting in

    If the a priori estimate of the process noise is zero, Then

    If the measurement noise is zero