Professional Documents
Culture Documents
• Limitations of linearization
1. Describe the Kalman filter as a two stage filter: (1) prediction and (2)
correction
PDF 1. Prediction
𝒩(xk̂ , Pk̂ )
not time!
𝒩(xk−1 ̂ )
̂ , Pk−1 𝒩(x̌k, P̌k) 𝒩(yk, Rk) Position (p)
Prediction
Measurement
Estimate at (k - 1) (Motion Model) Estimate at (k) (Observation Model)
e.g.,
e.g., GPS,
wheel odometry,
Lidar
inertial navigation
The Kalman Filter | Linear Dynamical System
• The Kalman Filter requires the following motion and measurement models:
2b Correction Prediction
x̌k (given motion model)
x̂k = x̌k + Kk(yk − Hkx̌k) at time k
PDF 1. Prediction
̂ )
̂ , Pk−1
𝒩(xk−1 𝒩(x̌k, P̌k) 𝒩(xk̂ , Pk̂ ) 𝒩(yk, Rk) Position (x)
[0 1 ] [Δt]
Our input is
1 Δt 0
PDF xk = xk−1 + uk−1 + wk−1 acceleration. This
could come from a
control system.
Position Observation
Position(p) This could be a
yk = [1 0] xk + vk GPS measurement.
Noise Densities
p
[ dt = p· ]
d 2p
x= dp u=a= 2 vk ∼ 𝒩(0, 0.05) wk ∼ 𝒩(0, (0.1)12×2)
dt
The Kalman Filter | Short Example
Data
[5] [ 0 1]
0 0.01 0
x̂0 ∼ 𝒩( , )
PDF
Δt = 0.5s
2 y1 = 2.2 [m]
Position(p) u0 = − 2 [m/s ]
The Kalman Filter | Short Example Solution
Prediction
[1.22]
0.88
=
[0.06 0.49]
0.04 0.06
=
Summary | The Kalman Filter
• The Kalman Filter is very similar to RLS but includes a motion model that tells
us how the state evolves over time