Unscented Kalman Filter

Hi all,

I found BFL while searching for a library to help me implement a Kalman filter on a nonlinear system--I'm pretty new to filtering in general, so if I'm misunderstanding anything here, please let me know.

The two libraries that seem most applicable to my project seem to be BFL and the bayes++ library (because this project is for work, I need a library that is either LGPL or MIT/BSD.)

Looking through the two libraries, I think BFL looks much better documented and easier to use, but I'm running into one issue:

I would really prefer to use an Unscented Kalman filter, rather than an EKF, because I wouldn't need to explicitly calculate the Jacobian of my system (which is generated at runtime, and I would really like to avoid programming up symbolic differentiation.) Also, I've read that the UKF is more accurate. Bayes++ does implement one, which is the only mark for it that I've seen so far...

I haven't been able to find a UKF implementation in this library. Is there one implemented that I've missed? If not, how painful would it be for me to implement one on top of this library?

Please correct me if I'm misunderstanding the advantages/disadvantages of UKF's over EKF's--like I said, I'm very new at this.

Thanks in advance,

Ben

Unscented Kalman Filter

Hi Ben,

[...]
> I found BFL while searching for a library to help me implement a Kalman filter on a nonlinear system--I'm pretty new to filtering in general, so if I'm misunderstanding anything here, please let me know.
>
> The two libraries that seem most applicable to my project seem to be BFL and the bayes++ library (because this project is for work, I need a library that is either LGPL or MIT/BSD.)
>
> Looking through the two libraries, I think BFL looks much better documented and easier to use, but I'm running into one issue:
>
> I would really prefer to use an Unscented Kalman filter, rather than an EKF, because I wouldn't need to explicitly calculate the Jacobian of my system (which is generated at runtime, and I would really like to avoid programming up symbolic differentiation.)  Also, I've read that the UKF is more accurate.  Bayes++ does implement one, which is the only mark for it that I've seen so far...
>
> I haven't been able to find a UKF implementation in this library.  Is there one implemented that I've missed?  If not, how painful would it be for me to implement one on top of this library?
>
> Please correct me if I'm misunderstanding the advantages/disadvantages of UKF's over EKF's--like I said, I'm very new at this.

I guess your questions were answered already in a previous thread?

Klaas

Unscented Kalman Filter

They were, thanks. This must be the post that got lost in the spam filter.

Ben

On Fri, Jan 29, 2010 at 6:33 AM, Klaas Gadeyne <klaas [dot] gadeyne [..] ...>wrote:

> Hi Ben,
>
> [...]
> > I found BFL while searching for a library to help me implement a Kalman
> filter on a nonlinear system--I'm pretty new to filtering in general, so if
> I'm misunderstanding anything here, please let me know.
> >
> > The two libraries that seem most applicable to my project seem to be BFL
> and the bayes++ library (because this project is for work, I need a library
> that is either LGPL or MIT/BSD.)
> >
> > Looking through the two libraries, I think BFL looks much better
> documented and easier to use, but I'm running into one issue:
> >
> > I would really prefer to use an Unscented Kalman filter, rather than an
> EKF, because I wouldn't need to explicitly calculate the Jacobian of my
> system (which is generated at runtime, and I would really like to avoid
> programming up symbolic differentiation.) Also, I've read that the UKF is
> more accurate. Bayes++ does implement one, which is the only mark for it
> that I've seen so far...
> >
> > I haven't been able to find a UKF implementation in this library. Is
> there one implemented that I've missed? If not, how painful would it be for
> me to implement one on top of this library?
> >
> > Please correct me if I'm misunderstanding the advantages/disadvantages of
> UKF's over EKF's--like I said, I'm very new at this.
>
> I guess your questions were answered already in a previous thread?
>
> Klaas
> --
> Orocos-Users mailing list
> Orocos-Users [..] ...
> http://lists.mech.kuleuven.be/mailman/listinfo/orocos-users
>