kalman filter upload

20
UNCLASSIFIED An ISO 9001:2008 Registered Company Kalman Filter Distribution Restrictions: < Enter any appropriate distribution restrictions in title master, (eg. Distribution D)> Data Rights: <Enter any applicable date rights restrictions, (eg. SBIR data rights or other similar information>; DP-FM-016, Rev 2 Effective Date: 22 February 2012

Upload: tushar-tank

Post on 02-Jul-2015

470 views

Category:

Documents


9 download

TRANSCRIPT

Page 1: Kalman filter upload

UN

CLA

SSIF

IED

An ISO 9001:2008 Registered Company

Kalman FilterDistribution Restrictions: < Enter any appropriate distribution restrictions in title master, (eg. Distribution D)>

Data Rights: <Enter any applicable date rights restrictions, (eg. SBIR data rights or other similar information>;

DP-FM-016, Rev 2

Effective Date: 22 February 2012

Page 2: Kalman filter upload

UN

CLA

SSIF

IED

Kalman Filter Facts

Dr. Rudolf Kalman is alive and well today (82 years old)

Important and used everywhere: GPS (predict update location), surface to air missiles (hit target), machine vision (track targets), brain computer interface

Not really a filter, it is an optimal estimator (infers parameters of interest from indirect, noise measurements)

It is recursive – so when a new measurement arrives it is processed and you get a new estimate

Performs Data Fusion usually between measured and estimated states

22

FEB

12

Notice: Use or disclosure of data contained on this sheet is subject to the restriction on the title page of this document.

2

Page 3: Kalman filter upload

UN

CLA

SSIF

IED

3

Conceptual Overview – Example Definition

Lost on the 1-dimensional line, boat is not moving

Imagine that you are guessing your position by looking at the stars using sextant

Position function of time: y(t)

Assume Gaussian distributed measurements (errors)

y

Page 4: Kalman filter upload

UN

CLA

SSIF

IED

4

Conceptual Overview - Prediction

Sextant Measurement at t1: Mean = z1 and Variance = z1

Optimal estimate of position is: ŷ(t1) = z1

Variance of error [y(t1) - ŷ(t1)] estimate:

2e (t1) = 2

z1

Boat in same position at time t2 - Predictedposition is z1

0 10 20 30 40 50 60 70 80 90 1000

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

z

pro

babili

ty ŷ(t1) = z1

Predicted Position

What if we also had a GPS unit?

Page 5: Kalman filter upload

UN

CLA

SSIF

IED

5

Conceptual Overview - Measurement

• So we have the prediction ŷ-(t2)

• GPS Measurement at t2: Mean = z2 and Variance = z2

• Need to correct the prediction by Sextant due to measurement to get ŷ(t2)

0 10 20 30 40 50 60 70 80 90 1000

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

prediction ŷ-(t2)

State (by looking

at the stars at t2)

Measurement

using GPS z(t2)

Page 6: Kalman filter upload

UN

CLA

SSIF

IED

6

0 10 20 30 40 50 60 70 80 90 1000

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

measurement

z(t2)

corrected optimal

estimate ŷ(t2)

prediction ŷ-(t2)

Conceptual Overview – Data Fusion

Kalman filter: fuse measurement and prediction based on confidence

Corrected mean is the new optimal estimate of position

New variance is smaller than either of the previous two variances

What if the boat is moving?

Page 7: Kalman filter upload

UN

CLA

SSIF

IED

7

Conceptual Overview – Prediction Model

0 10 20 30 40 50 60 70 80 90 1000

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

ŷ(t2)

Naïve Prediction

(sextant) ŷ-(t3)

At time t3, boat moves with velocity dy/dt=u

Naïve approach: Shift probability to the right to predict

This would work if we knew the velocity exactly (perfect model)Try and predict where

it winds up.

Page 8: Kalman filter upload

UN

CLA

SSIF

IED

8

Conceptual Overview – Prediction Model

But you may not be so sure about the exact velocity

Better to assume imperfect model by adding Gaussian noise

dy/dt = u + w

Distribution for prediction moves and spreads out

Assumptions: prediction is linear, noise is Gaussian

0 10 20 30 40 50 60 70 80 90 1000

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

Prediction ŷ-(t3)

ŷ(t2)

Naïve Prediction

(sextant) ŷ-(t3)

Page 9: Kalman filter upload

UN

CLA

SSIF

IED

9

0 10 20 30 40 50 60 70 80 90 1000

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

Prediction ŷ-(t3) Sextant

Measurement z(t3) GPS

Corrected optimal estimate ŷ(t3)

Updated Sextant position using

GPS

• Now we take a measurement (GPS) at t3

• Need to once again correct the prediction (fusion)

• Recursive – rinse and repeat as time goes on

Conceptual Overview – Update

Update, recursively

Page 10: Kalman filter upload

UN

CLA

SSIF

IED

10

Conceptual Overview

Optimal estimator only if: Prediction model is linear (function of measurements)

All error (noise) is Gaussian: model error, measurement error

Why is Kalman Filter so popular Good results in practice due to optimality and structure.

Convenient form for online real time processing.

Easy to formulate and implement given a basic understanding.

Measurement equations need not be inverted.

Page 11: Kalman filter upload

UN

CLA

SSIF

IED

State Space Equations

22

FEB

12

Notice: Use or disclosure of data contained on this sheet is subject to the restriction on the title page of this document.

11

EstimatedState(now)

ObservedMeasurement

EstimatedState

(before)

ControlInput

How do you find A,B,H?AWGN?

Page 12: Kalman filter upload

UN

CLA

SSIF

IED

Update Equations

22

FEB

12

Notice: Use or disclosure of data contained on this sheet is subject to the restriction on the title page of this document.

12

Description Equation

State Prediction Where do we end up

Covariance Prediction When we get there, how much error

InnovationCompare Reality to Prediction

Innovation CovarianceCompare real error to predicted error

Kalman GainWhat do you trust more?

State UpdateNew estimate of where we are

Covariance UpdateNew estimate of error

Input

Output

Place holder

Page 13: Kalman filter upload

UN

CLA

SSIF

IED

Algorithm

22

FEB

12

Notice: Use or disclosure of data contained on this sheet is subject to the restriction on the title page of this document.

13

Prediction (Time Update)

(1) Project the state ahead

(2) Project the error covariance ahead

Correction (Measurement Update)

(1) Compute the Kalman Gain

(2) Update estimate with measurement zk

(3) Update Error Covariance

Page 14: Kalman filter upload

UN

CLA

SSIF

IED

Measuring Constant Voltage (Classic Example 1)

22

FEB

12

Notice: Use or disclosure of data contained on this sheet is subject to the restriction on the title page of this document.

14

Page 15: Kalman filter upload

UN

CLA

SSIF

IED

Predicting Trajectory of Projectile (Angry Bird)

22

FEB

12

Notice: Use or disclosure of data contained on this sheet is subject to the restriction on the title page of this document.

15

Page 16: Kalman filter upload

UN

CLA

SSIF

IED

Equations

22

FEB

12

Notice: Use or disclosure of data contained on this sheet is subject to the restriction on the title page of this document.

16

Page 17: Kalman filter upload

UN

CLA

SSIF

IED

Simulation Results

22

FEB

12

Notice: Use or disclosure of data contained on this sheet is subject to the restriction on the title page of this document.

17

Page 18: Kalman filter upload

UN

CLA

SSIF

IED

Modified TWS example

22

FEB

12

Notice: Use or disclosure of data contained on this sheet is subject to the restriction on the title page of this document.

18

*][:

},,,{:

wwEQCov

yxyxState y

Page 19: Kalman filter upload

UN

CLA

SSIF

IED

Derivation

22

FEB

12

Notice: Use or disclosure of data contained on this sheet is subject to the restriction on the title page of this document.

19

Page 20: Kalman filter upload

UN

CLA

SSIF

IED

What if Assumptions don’t hold

22

FEB

12

Notice: Use or disclosure of data contained on this sheet is subject to the restriction on the title page of this document.

20