[ODE] ODE spatial inconsistency

Marcin Pilat pilat at cpsc.ucalgary.ca
Thu Jun 21 13:57:29 MST 2007


Thanks STenyaK,

That does make sense and the reason why ODE doesn't actually fix this  
is that the user can specify any range of values that the simulation  
can run in.  For my simulation, if I know that my world will be say  
1000 units by 1000 units, I can calculate how many decimal places I  
need to get the proper # of digits (say for double precision) and  
round everything off to that much so that everything is done  
consistently.

Any other suggestions for this?

Marcin


On 21-Jun-07, at 12:27 PM, STenyaK (Bruno González) wrote:

> On 6/21/07, Marcin Pilat <pilat at cpsc.ucalgary.ca> wrote:
>> deterministic version of ODE (using dWorldStep) and when I run the
>> sample at the same x,y position, the result is identical every run.
>> However, the problem arises when I vary that x,y position.
>
> That makes sense if you use a "lossy" format such as IEEE754, which is
> the usual thing in PCs. Imagine you model a screw of 1mm in planet
> earth and another one 10.000 astronomic units far away... you can't
> pretend both to behave exactly the same, up to an infinite level of
> precision, given current computers hardware. You probably can't even
> model that without errors of *meters*... the furthest screw would
> probably have a size of either 0m or... 1m (or whatever). That's an
> extreme case, but should be valid for smaller numbers too.
>
> If you want extreme precission, i'm sure there are math libraries that
> provide classes with variable precision. For example, allowing to use
> 128bits for integers instead of 32bits when you need it (at the
> expense of processing time); that might be what you're looking for.
>
> All the above is AFAIK, someone correct me if i'm wrong.
> -- 
> Saludos,
>     STenyaK
>
> _______________________________________________
> Msn/Jabber: stenyak AT gmail.com
> ICQ: 153709484
> Webs: http://1ksurvivor.homeip.net 1kSurvivor
> http://motorsport-sim.org Motorsport
> http://emuletutorial.info EmuleTutorial




More information about the ODE mailing list