Reply by Vladimir Vassilevsky●December 21, 20092009-12-21
Tim Wescott wrote:
> On Mon, 21 Dec 2009 16:57:10 -0500, Jerry Avins wrote:
>
>
>>fitlike min wrote:
>>
>>>On Dec 22, 2:52 am, "hagai_sela" <hagai.s...@gmail.com> wrote:
>>>
>>>>Hi,
>>>>I am trying to implement an unscented kalman filter, as described
>>>>here:
>>>>
>>>>
>>>
>>>I have never smelt a Kalman filter before but understood that it not
>>>smell of oil.
>>
>>It's a strange name, but it's real.
>>http://en.wikipedia.org/wiki/Kalman_filter#Unscented_Kalman_filter
>
>
> If you try applying the extended Kalman filter to a problem that's
> nonlinear enough, you may well find yourself exclaiming "Man, this
> stinks!". And then you'll find a solution in the unscented Kalman
> (maybe).
>
> I suspect it was an effort by mathematicians to one-up the physicists who
> came up with strange and charm quarks.
They have to use fancy words to get financing. "Low density parity code"
is such a lousy brand. "TURBO code" sounds great.
VLV
Reply by Tim Wescott●December 21, 20092009-12-21
On Mon, 21 Dec 2009 16:57:10 -0500, Jerry Avins wrote:
> fitlike min wrote:
>> On Dec 22, 2:52 am, "hagai_sela" <hagai.s...@gmail.com> wrote:
>>> Hi,
>>> I am trying to implement an unscented kalman filter, as described
>>> here:
>>>
>>>
>>
>> I have never smelt a Kalman filter before but understood that it not
>> smell of oil.
>
> It's a strange name, but it's real.
> http://en.wikipedia.org/wiki/Kalman_filter#Unscented_Kalman_filter
If you try applying the extended Kalman filter to a problem that's
nonlinear enough, you may well find yourself exclaiming "Man, this
stinks!". And then you'll find a solution in the unscented Kalman
(maybe).
I suspect it was an effort by mathematicians to one-up the physicists who
came up with strange and charm quarks.
--
www.wescottdesign.com
Reply by Jerry Avins●December 21, 20092009-12-21
fitlike min wrote:
> On Dec 22, 2:52 am, "hagai_sela" <hagai.s...@gmail.com> wrote:
>> Hi,
>> I am trying to implement an unscented kalman filter, as described here:
>>
>
>
> I have never smelt a Kalman filter before but understood that it not
> smell of oil.
It's a strange name, but it's real.
http://en.wikipedia.org/wiki/Kalman_filter#Unscented_Kalman_filter
Jerry
--
Engineering is the art of making what you want from things you can get.
�����������������������������������������������������������������������
Reply by fitlike min●December 21, 20092009-12-21
On Dec 22, 2:52�am, "hagai_sela" <hagai.s...@gmail.com> wrote:
> Hi,
> I am trying to implement an unscented kalman filter, as described here:
>
I have never smelt a Kalman filter before but understood that it not
smell of oil.
Reply by Tim Wescott●December 21, 20092009-12-21
On Mon, 21 Dec 2009 07:52:40 -0600, hagai_sela wrote:
> Hi,
> I am trying to implement an unscented kalman filter, as described here:
>
> http://cslu.cse.ogi.edu/nsel/ukf/node6.html
>
> My problem is with equation 18 - What does the multiplication mean? As
> far as I understand, the first (Yi - y^) is a row vector, and the second
> one is a column vector, therefore this is a dot product. But then the
> result of the expression inside the sigma is a scalar, and Pyy is
> supposed to be a matrix.
> I also tried to interpret this as a vector direct product (column vector
> multiplied by a row vector), the result is a matrix but I get bad
> results when trying to update the covariance matrix.
>
> Help, please...
>
> Thanks,
> Hagai.
It's a column vector times a row vector, and the result is a matrix.
Basically, you're starting from the definition of the covariance matrix
-- compare (18) with the 'formal' definition of the covariance matrix in
a book on Kalman filtering -- but you're only computing it for a few
discrete y_i instead of the space of all possible y_i.
Bad results when trying to update the covariance matrix can come from
many things; two on the edges of the problem space are that (a) the UKF
isn't guaranteed to work for any problem, and (b) you may have a garden-
variety bug in your code. There's a whole bunch of other possibilities
between and/or around these, but yours is probably a mix of the two.
If you get desperate, try the EKF for a simple linear system -- you
should be getting the same results for P_y as you get from the (much
simpler) computation for the Kalman.
--
www.wescottdesign.com
Reply by hagai_sela●December 21, 20092009-12-21
Hi,
I am trying to implement an unscented kalman filter, as described here:
http://cslu.cse.ogi.edu/nsel/ukf/node6.html
My problem is with equation 18 - What does the multiplication mean? As far
as I understand, the first (Yi - y^) is a row vector, and the second one is
a column vector, therefore this is a dot product. But then the result of
the expression inside the sigma is a scalar, and Pyy is supposed to be a
matrix.
I also tried to interpret this as a vector direct product (column vector
multiplied by a row vector), the result is a matrix but I get bad results
when trying to update the covariance matrix.
Help, please...
Thanks,
Hagai.