Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss
Implementation of UKF for Localization?
35 views
Skip to first unread message
AO
Mar 26, 2012, 2:25:13 PM3/26/12
to
Hi,
I have tried to implement Unscented Kalman Filters for Localization using the algorithm given in Probabilistic Robotics (Thrun, Burgaard, Fox). I don't know if this is the right forum to ask such a question. If you would be kind enough to point me to the right location that would be great.
http://mason.gmu.edu/~asiddiq5/kf.html (contains EKF, UKF code, analysis, data and results)
I am using UKF for localization algorithm (on page 221 and table 7.4) with some sample data. (sorry working on reproducing the algorithm in electronic format).
My question is :
1) if I have more than one landmark measurements do I loop over lines 10-16 for table 7.4? if you see, this eventuality (where there are more than one landmark measurements) is handled for ekfs but this is not considered in the algorithm for UKFs (in table 7.4).
2) when calculating the sigma points I am forced to use sqrt instead of cholesky factorization for calculating the square root term in equations 3.66 (on page 65). Is this ok? When using cholesky factorization I am getting the following error:
Here is the complete error :
??? Error using ==> chol Matrix must be positive definite.
Error in ==> SigmaPoints at 10 U = chol((n+lambda)*covariance);
It goes fine with the first iteration when the covariance is a diagonal matrix. At the very next call to the sigmapts the covariance of the state is no longer diagonal and that is when it gives an error.
Perhaps another problem is that the state (x,y,theta ) is already incorrect by then (meaning x is already in the negative or has veered to the left). p.s. someone told me that the covariance has to be positive definite but I didnt know what to do with that.
I tried implementing EKF version of it and the trajectory seems to be pretty close to the expected trajectory even though the covariance matrices are shrinking and then vanishing.
Any help would be appreciated.
************** comp.robotics.research (moderated) **************
Summary: Academic, government & industry research in robotics.
Charter and information: http://www.metamech.com/crr
Meta-discussions/information: crr-r...@metamech.com
I have tried to implement Unscented Kalman Filters for Localization using the algorithm given in Probabilistic Robotics (Thrun, Burgaard, Fox). I don't know if this is the right forum to ask such a question. If you would be kind enough to point me to the right location that would be great.
http://mason.gmu.edu/~asiddiq5/kf.html (contains EKF, UKF code, analysis, data and results)
I am using UKF for localization algorithm (on page 221 and table 7.4) with some sample data. (sorry working on reproducing the algorithm in electronic format).
My question is :
1) if I have more than one landmark measurements do I loop over lines 10-16 for table 7.4? if you see, this eventuality (where there are more than one landmark measurements) is handled for ekfs but this is not considered in the algorithm for UKFs (in table 7.4).
2) when calculating the sigma points I am forced to use sqrt instead of cholesky factorization for calculating the square root term in equations 3.66 (on page 65). Is this ok? When using cholesky factorization I am getting the following error:
Here is the complete error :
??? Error using ==> chol Matrix must be positive definite.
Error in ==> SigmaPoints at 10 U = chol((n+lambda)*covariance);
It goes fine with the first iteration when the covariance is a diagonal matrix. At the very next call to the sigmapts the covariance of the state is no longer diagonal and that is when it gives an error.
Perhaps another problem is that the state (x,y,theta ) is already incorrect by then (meaning x is already in the negative or has veered to the left). p.s. someone told me that the covariance has to be positive definite but I didnt know what to do with that.
I tried implementing EKF version of it and the trajectory seems to be pretty close to the expected trajectory even though the covariance matrices are shrinking and then vanishing.
Any help would be appreciated.
************** comp.robotics.research (moderated) **************
Summary: Academic, government & industry research in robotics.
Charter and information: http://www.metamech.com/crr
Meta-discussions/information: crr-r...@metamech.com
0 new messages