Last active
May 27, 2022 04:26
-
-
Save aztennenbaum/3a8be5c5c8b9d536ee756f10e53e777b to your computer and use it in GitHub Desktop.
Baysian derivation of kalman filter
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Theorems: | |
1. Linear transform of a normal random variable (matrix cookbook, 8.1.4): | |
Assume: | |
X ~ N(x,Px), | |
f(v) = A*v | |
Then: | |
f(X) ~ N(A*x,A*Px*A') | |
2. Conditional Distribution (matrix cookbook, 8.1.3): | |
[X;Y]~N([x0;y_expected],[Pxx,Pxy;Pxy';Pyy]); | |
p(x1|y_observed) ~ N(x1,Pxx_new)) | |
K = Pxy*(Pyy)^-1 | |
x1 = x0-K*(y_expected-y_observed) | |
Pxx_new = Pxx-K*(Pxy') | |
We have discrete time linear system: | |
[X1;Y0]=[A,B;C,D]*[X0;R] | |
Random variables: | |
Uncertainty: X0~N(x0,P0) | |
Noise: R~N(0,I) | |
Uncorrelated, so we can say: | |
[X0;R]~N([x0;0],[P0,0;0,I])] | |
Apply theorem 1: | |
f(v)=[A,B;C,D]*v | |
[X1;Y0]~f([X0;R]) | |
[X1;Y0]~N([A,B;C,D]*[x0,0],[A,B;C,D]*[P0,0;0,I]*[A,B;C,D]'); | |
[X1;Y0]~N([A*x0;C*x0],[A*P0*A'+B*B',A*P0*C'+B*D'; | |
C*P0*A'+D*B',C*P0*C'+D*D']); | |
Apply theorem 2: | |
p(x1|y0) = N(x1,P1) | |
K = (A*P0*C'+B*D')*(C*P0*C'+D*D')^-1 | |
x1 = A*x0 - K*(C*x0 - y0) | |
P1 = A*P0*A'+B*B - K*(C*P0*A'+D*B') | |
For seperate update and propagate: | |
Update: | |
A=I | |
B=0 | |
K = (P0*C')*(C*P0*C'+D*D')^-1 | |
P(x1|y0) ~ N(x0 - K*(C*x0 - y0), P0 - K*(C*P0)) | |
Propagate: | |
P(x1|x0) ~ N(A*x0, A*P0*A'+B*B') | |
References: | |
Matrix cookbook - https://www.math.uwaterloo.ca/~hwolkowi/matrixcookbook.pdf | |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment