[Rock-dev] Representation of invalid positions vs. unknown uncertainty
Leif Christensen
leif.christensen at dfki.de
Thu Jan 19 19:32:38 CET 2012
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Hi,
some thoughts on that:
With respect to IEEE 754 floating point conventions, using NaN for
unknown uncertainty is an option, since it is commonly used for
undefined, but also uninitialized or even unknown values. Con is, that
operations with NaN may throw an error if honoring singalling NaNs is
on (-fsignaling-nans). On most system the default will be quiet NaNs,
that will propagate through your operations.
Concerning 0 vs. Inf in the off-diagonal positions, there should be no
difference at least for the case multiplying with a vector:
case 1:
Inf 0 0 a aInf + b0 + c0
0 Inf 0 x b = a0 + bInf + c0
0 0 Inf c a0 + b0 + cInf
Inf
= Inf
Inf
case 2:
Inf Inf Inf a aInf + bInf + cInf
Inf Inf Inf x b = aInf + bInf + cInf
Inf Inf Inf c aInf + bInf + cInf
Inf
= Inf
Inf
since Inf + Inf is defined as Inf.
If a or b or c is zero, the resulting row will be NaN for both cases.
Greetings,
Leif
P.S. Hm, just checked 0 multiplied by Inf which gives on my system
- -NaN and not NaN, using gcc -std=gnu99. Weird
On 19.01.2012 16:18, Sylvain Joyeux wrote:
> On 01/19/2012 09:48 AM, Jakob Schwendner wrote:
>>> Which is the default value set by RigidBodyState::invalidate()
>>> and marks the position as unknown (infinite uncertainty on the
>>> position).
>> I would also think that the NaNs in the off-diagonal are
>> misplaced if it wants to say infinite uncertainty. Should be 0 in
>> my opinion.
> That's something I meant to ask ...
>
> What would
>
> Inf 0 0 0 Inf 0 0 0 Inf
>
> actually mean (if it does mean something). What would be the
> difference with:
>
> Inf Inf Inf Inf Inf Inf Inf Inf Inf
>
> Sylvain _______________________________________________ Rock-dev
> mailing list Rock-dev at dfki.de
> http://www.dfki.de/mailman/cgi-bin/listinfo/rock-dev
- --
!!! ATTENTION! NEW TELEPHONE NUMBER! !!!
Leif Christensen --- Underwater Robotics
DFKI Bremen
Robotics Innovation Center
Robert-Hooke-Straße 5
28359 Bremen, Germany
Phone: +49 (0)421 17845-4149
Fax: +49 (0)421 17845-4150
E-Mail: leif.christensen at dfki.de
Weitere Informationen: http://www.dfki.de/robotik
-----------------------------------------------------------------------
Deutsches Forschungszentrum fuer Kuenstliche Intelligenz GmbH
Firmensitz: Trippstadter Straße 122, D-67663 Kaiserslautern
Geschaeftsfuehrung: Prof. Dr. Dr. h.c. mult. Wolfgang Wahlster
(Vorsitzender) Dr. Walter Olthoff
Vorsitzender des Aufsichtsrats: Prof. Dr. h.c. Hans A. Aukes
Amtsgericht Kaiserslautern, HRB 2313
Sitz der Gesellschaft: Kaiserslautern (HRB 2313)
USt-Id.Nr.: DE 148646973
Steuernummer: 19/673/0060/3
-----------------------------------------------------------------------
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/
iEYEARECAAYFAk8YYcYACgkQgrIQzq7NbhFtcwCgrkWKZN1uJ1sl0Wkf7+gM+9bi
/I4AoLmB/O6GgTKrC9nJQqNCL+MYAsZf
=TBUc
-----END PGP SIGNATURE-----
More information about the Rock-dev
mailing list