NCAF Manchester 2000 1 6 July 2000 Graham Hesketh Information Engineering Group Rolls-Royce Strategic Research Centre.

Slides:



Advertisements
Similar presentations
Probabilistic Reasoning over Time
Advertisements

EKF, UKF TexPoint fonts used in EMF.
Use of Kalman filters in time and frequency analysis John Davis 1st May 2011.
AGC DSP AGC DSP Professor A G Constantinides©1 Modern Spectral Estimation Modern Spectral Estimation is based on a priori assumptions on the manner, the.
Extended Kalman Filter (EKF) And some other useful Kalman stuff!
ECE 8443 – Pattern Recognition ECE 8423 – Adaptive Signal Processing Objectives: Newton’s Method Application to LMS Recursive Least Squares Exponentially-Weighted.
Observers and Kalman Filters
Introduction to Mobile Robotics Bayes Filter Implementations Gaussian filters.
Motion Tracking. Image Processing and Computer Vision: 82 Introduction Finding how objects have moved in an image sequence Movement in space Movement.
280 SYSTEM IDENTIFICATION The System Identification Problem is to estimate a model of a system based on input-output data. Basic Configuration continuous.
Introduction to Kalman Filter and SLAM Ting-Wei Hsu 08/10/30.
Development of Empirical Models From Process Data
Discriminative Training of Kalman Filters P. Abbeel, A. Coates, M
Probabilistic Robotics
Tracking a maneuvering object in a noisy environment using IMMPDAF By: Igor Tolchinsky Alexander Levin Supervisor: Daniel Sigalov Spring 2006.
Estimation and the Kalman Filter David Johnson. The Mean of a Discrete Distribution “I have more legs than average”
Course AE4-T40 Lecture 5: Control Apllication
Probabilistic Robotics Bayes Filter Implementations Gaussian filters.
Kalman Filtering Pieter Abbeel UC Berkeley EECS Many slides adapted from Thrun, Burgard and Fox, Probabilistic Robotics TexPoint fonts used in EMF. Read.
Overview and Mathematics Bjoern Griesbach
Adaptive Signal Processing
ROBOT MAPPING AND EKF SLAM
Principles of the Global Positioning System Lecture 13 Prof. Thomas Herring Room A;
Kalman filter and SLAM problem
Muhammad Moeen YaqoobPage 1 Moment-Matching Trackers for Difficult Targets Muhammad Moeen Yaqoob Supervisor: Professor Richard Vinter.
Colorado Center for Astrodynamics Research The University of Colorado STATISTICAL ORBIT DETERMINATION Project Report Unscented kalman Filter Information.
Introduction to estimation theory Seoul Nat’l Univ.
Colorado Center for Astrodynamics Research The University of Colorado 1 STATISTICAL ORBIT DETERMINATION Satellite Tracking Example of SNC and DMC ASEN.
Computer vision: models, learning and inference Chapter 19 Temporal models.
Computer Vision Group Prof. Daniel Cremers Autonomous Navigation for Flying Robots Lecture 6.2: Kalman Filter Jürgen Sturm Technische Universität München.
Computer vision: models, learning and inference Chapter 19 Temporal models.
Probabilistic Robotics Robot Localization. 2 Localization Given Map of the environment. Sequence of sensor measurements. Wanted Estimate of the robot’s.
The Kalman Filter ECE 7251: Spring 2004 Lecture 17 2/16/04
Kalman Filter (Thu) Joon Shik Kim Computational Models of Intelligence.
Computer Vision - A Modern Approach Set: Tracking Slides by D.A. Forsyth The three main issues in tracking.
Complete Pose Determination for Low Altitude Unmanned Aerial Vehicle Using Stereo Vision Luke K. Wang, Shan-Chih Hsieh, Eden C.-W. Hsueh 1 Fei-Bin Hsaio.
Probabilistic Robotics Bayes Filter Implementations Gaussian filters.
Young Ki Baik, Computer Vision Lab.
Karman filter and attitude estimation Lin Zhong ELEC424, Fall 2010.
CHAPTER 4 Adaptive Tapped-delay-line Filters Using the Least Squares Adaptive Filtering.
HQ U.S. Air Force Academy I n t e g r i t y - S e r v i c e - E x c e l l e n c e Improving the Performance of Out-of-Order Sigma-Point Kalman Filters.
University of Colorado Boulder ASEN 5070: Statistical Orbit Determination I Fall 2014 Professor Brandon A. Jones Lecture 18: Minimum Variance Estimator.
The “ ” Paige in Kalman Filtering K. E. Schubert.
Processing Sequential Sensor Data The “John Krumm perspective” Thomas Plötz November 29 th, 2011.
Unscented Kalman Filter 1. 2 Linearization via Unscented Transform EKF UKF.
An Introduction To The Kalman Filter By, Santhosh Kumar.
Tracking with dynamics
By: Aaron Dyreson Supervising Professor: Dr. Ioannis Schizas
Extended Kalman Filter
Nonlinear State Estimation
September 28, 2000 Improved Simultaneous Data Reconciliation, Bias Detection and Identification Using Mixed Integer Optimization Methods Presented by:
Colorado Center for Astrodynamics Research The University of Colorado 1 STATISTICAL ORBIT DETERMINATION Kalman Filter with Process Noise Gauss- Markov.
The Unscented Kalman Filter for Nonlinear Estimation Young Ki Baik.
Robust Localization Kalman Filter & LADAR Scans
State-Space Recursive Least Squares with Adaptive Memory College of Electrical & Mechanical Engineering National University of Sciences & Technology (NUST)
DSP-CIS Part-III : Optimal & Adaptive Filters Chapter-9 : Kalman Filters Marc Moonen Dept. E.E./ESAT-STADIUS, KU Leuven
Least Squares Measurement model Weighted LSQ  Optimal estimates  Linear  Unbiased  Minimum variance.
Probabilistic Robotics Bayes Filter Implementations Gaussian filters.
STATISTICAL ORBIT DETERMINATION Kalman (sequential) filter
Using Sensor Data Effectively
STATISTICAL ORBIT DETERMINATION Coordinate Systems and Time Kalman Filtering ASEN 5070 LECTURE 21 10/16/09.
PSG College of Technology
Unscented Kalman Filter
Unscented Kalman Filter
Kalman Filtering: Control with Limited/Noisy Measurements
Unscented Kalman Filter
Bayes and Kalman Filter
Kalman Filtering COS 323.
Principles of the Global Positioning System Lecture 13
Kalman Filter: Bayes Interpretation
Presentation transcript:

NCAF Manchester July 2000 Graham Hesketh Information Engineering Group Rolls-Royce Strategic Research Centre

NCAF Manchester July 2000 Basic Kalman Filter Equations Prediction Updating (correction)

NCAF Manchester July 2000 Definitions System state: mean estimate x - vector of size q e.g. for a single channel, constant acceleration model x 1 = position x 2 = velocity x 3 = acceleration

NCAF Manchester July 2000 Definitions System state: covariance (uncertainty) P - matrix of size q x q e.g. P 11 = uncertainty (variance) in position P 22 = uncertainty in velocity P 12 = covariance in position/velocity...

NCAF Manchester July 2000 Definitions State dynamics: state transition matrix A - matrix of size q x q e.g. for the constant acceleration example

NCAF Manchester July 2000 Definitions State dynamics: process noise (uncertainty) Q - matrix of size q x q e.g. (modelling acceleration as constant with a long duration; 3 rd derivative is random and instantaneous)

NCAF Manchester July 2000 Definitions Measurement space: observation matrix C - matrix of size r x q e.g. for the simple single measurement constant acceleration model: C = [1 0 0]

NCAF Manchester July 2000 Definitions Measurement space: innovation sequence e - error between observation and prediction a.k.a. the innovation sequence should be zero-mean, random white noise  if not, there is a problem with either » the model, or » a sensor

NCAF Manchester July 2000 Definitions Measurement space: measurement noise matrix R - matrix of size r x r e.g. for the single measurement case, R =  2 R is the covariance matrix of the measurement noise. It is usually diagonal (i.e. uncorrelated sensor noise).

NCAF Manchester July 2000 Definitions Correction phase: Kalman gain matrix K - matrix of size q x r K is the optimal correction factor to obtain the Minimum Mean Squared Error estimate of the system state mean and covariance

NCAF Manchester July 2000 Definitions Confidence: Relative contributions to uncertainty Predicted measurement - uncertainty due to model Actual measurement - uncertainty due to sensor noise Relative weighting for the innovation error

NCAF Manchester July 2000 Definitions Correction phase: system stiffness Model uncertainty >> Measurement noise  Believe Sensor (e.g. Gain is high) Low system stiffness, low lag Model uncertainty << Measurement noise  Believe Model (e.g. Gain is low) High system stiffness, high lag Relative weighting (Gain) for the innovation error

NCAF Manchester July 2000 Definitions Correction phase: updating state estimates “Best” estimates of the system state mean and covariance, at time k, conditioned on all measurements up to and including time k

NCAF Manchester July 2000 Comparison with Fixed-Coefficient Filters If the A, C, Q and R matrices do not vary with k then the single channel CA Kalman filter settles down have steady-state Kalman gains which are equivalent to a standard  -filter, but with optimum coefficients selected automatically. This approach is known as Weiner filtering. It is computationally much simpler than a Kalman filter, but the steady-state assumptions may not be valid in many practical cases.

NCAF Manchester July 2000 Advantages of Kalman Filters Copes with variable A, C, Q and R matrices Copes with the large uncertainty of the initialisation phase Copes with missing data Provides a convenient measure of estimation accuracy (via the covariance matrix P) Fuses information from multiple-sensors

NCAF Manchester July 2000 Disadvantages of Kalman Filters Computationally complex (especially for large numbers of sensor channels r) Requires conditional independence of the measurement errors, sample-to-sample Requires linear models for state dynamics and observation processes (A and C) Getting a suitable value for Q (a.k.a. tuning) can be something of a black art

NCAF Manchester July 2000 Dealing with the disadvantages [1] Computationally complex (especially for large numbers of sensor channels r) Computers are getting faster all the time New algorithms for fast matrix inversion

NCAF Manchester July 2000 Dealing with the disadvantages [2] Requires conditional independence of the measurement errors, sample-to-sample The independence criterion can be removed by using Covariance Intersection updating A B C a,b,c

NCAF Manchester July 2000 Covariance Intersection vs Kalman updates Standard Kalman filter update: Covariance Intersection update: Kalman filter covariance (alternative form):

NCAF Manchester July 2000 Dealing with the disadvantages [3] Requires linear models for state dynamics and observation processes (A and C) This can be overcome by using the EKF (Extended Kalman Filter), but only at the expense of your sanity and all your free time A superior alternative to EKF is the Unscented Filter

NCAF Manchester July 2000 Unscented Filter - General Problem n-dimensional vector random variable x with mean and covariance We want to predict the distribution of an m-dimensional vector random variable y, where y is related to x by the non-linear transformation In filtering there are two such transformations: » the state transition » the observation

NCAF Manchester July 2000 Unscented Filter - Solution Compute the set  of 2n points from the rows or columns of the matrices This set is zero mean with covariance P. Compute a set of points with the same covariance, but with mean as Transform each point as Compute and by computing the mean and covariance of the 2n points in the set This is computationally much simpler and significantly more accurate (lower MSE) than the corresponding EKF

NCAF Manchester July 2000 Dealing with the disadvantages [4] Getting a suitable value for Q (a.k.a. tuning) can be something of a black art There are many heuristics for dealing with process noise (mainly to keep the filter stable); one method is to switch between alternative models We have Matlab toolboxes for learning Q (and all the other matrices) from empirical data, using a technique called Expectation Maximisation (EM)

NCAF Manchester July 2000 Summary The Kalman Filter is the way to fuse multiple data sources, providing that their errors are conditionally independent Covariance Intersection is the conservative way to deal with inter-dependence in distributed systems The Unscented Filter is universally superior to the EKF for all cases involving non-linear transformations Many heuristics exist to extend and improve the basic Kalman filter for tracking and sensor bias detection

NCAF Manchester July 2000 End of Talk Unless there are any questions – make ‘em brief