On Wed, May 19, 2010 at 9:06 PM, <philip [dot] dutoit [..] ...> wrote:

[forgot to answer your second question]

> Also, the ASIR_filter option is commented out in the code.  Does the ASIR_filter not work?

I would challenge you to check it and submit a patch if it doesn't :-)

Klaas
_______________________________________________
I hereby promise not to top-post on the
BFL mailing list
BFL [..] ...
http://lists.mech.kuleuven.be/mailman/listinfo/bfl

Hi Philippe,

I only had a quick glance, so I might be _completely_ wrong here. But
since I don't want Tinne to loose focus during her
PhD-presentation-preparation sprint ;-), I'm gonna give it a try...

On Wed, May 19, 2010 at 9:06 PM, <philip [dot] dutoit [..] ...> wrote:
> Thank you for an excellent and well-documented library.
>
> Running bfl/build/examples/compare_filters/test_compare_filters with the EK_particle_filter argument yields the following runtime error:
>
> $ ./test_compare_filters EK_particle_filter
> ==================================================
> Test of switching between different filters
> Mobile robot localisation example
> ==================================================
> Measurment model H = [1,3]((0,-2,0))
> Prior initialised
> Prior Mean =
> [3](-0.1,0.1,0.785398)
> Covariance =
> [3,3]((0.04,0,0),(0,0.04,0),(0,0,0.154213))
> Mobile robot initialised
> Using the Extended Particle Kalman Filter, 1000 samples, dynamic resampling
> MAIN: Starting estimation
> Assertion failed: (MeasModel != NULL), function MeasurementModelSet, file bfl/src/pdf/filterproposaldensity.cpp, line 84.
> Abort trap
> -----------------------------------------------------------------
>
> Am I supposed to provide a measurement model?

Yes and no :-)
As you probably know, A EK Particle Filter, is an "advanced" particle
filter, that uses the MeasurementModel in its proposal step, in order
to generate samples where "they should be" and hence improve the
quality of the filter. Therefore, you really need the
measurementmodel.

However, by default (and it's ugly legacy), the test compare example
does deadreckoning
[kgad@ampere /kgad/workspaces/orocos]$
wcgrep -r USE_MEASUREMENTS
./bfl/examples/compare_filters/test_compare_filters.cpp:303: if
(USE_MEASUREMENTS)
./bfl/examples/mobile_robot_wall_cts.h:26:#define USE_MEASUREMENTS 0

This is a nice illustration of the fact that we probably also need
some exhaustive testing of the examples :-(

Now, when I modified this define, in my setup (debian testing, bfl
trunk, boost as matrix/library) don't get the original exception
anymore, but get the following boost assert

Check failed in file /usr/include/boost/numeric/ublas/storage.hpp at line 195:
i < size_
terminate called after throwing an instance of
'boost::numeric::ublas::bad_index'
what(): bad index
Aborted

I didn't look any further into this one.

HTH and HYRV (your result varies :-)

Klaas
_______________________________________________
I hereby promise not to top-post on the
BFL mailing list
BFL [..] ...
http://lists.mech.kuleuven.be/mailman/listinfo/bfl

Thank you for an excellent and well-documented library.

Running bfl/build/examples/compare_filters/test_compare_filters with the EK_particle_filter argument yields the following runtime error:

$ ./test_compare_filters EK_particle_filter
==================================================
Test of switching between different filters
Mobile robot localisation example
==================================================
Measurment model H = [1,3]((0,-2,0))
Prior initialised
Prior Mean =
[3](-0.1,0.1,0.785398)
Covariance =
[3,3]((0.04,0,0),(0,0.04,0),(0,0,0.154213))
Mobile robot initialised
Using the Extended Particle Kalman Filter, 1000 samples, dynamic resampling
MAIN: Starting estimation
Assertion failed: (MeasModel != NULL), function MeasurementModelSet, file bfl/src/pdf/filterproposaldensity.cpp, line 84.
Abort trap
-----------------------------------------------------------------

Am I supposed to provide a measurement model?

Also, the ASIR_filter option is commented out in the code. Does the ASIR_filter not work?

Thanks.

_______________________________________________
I hereby promise not to top-post on the
BFL mailing list
BFL [..] ...
http://lists.mech.kuleuven.be/mailman/listinfo/bfl