[ODE] Large mass and accuracy

[Bob Dowland]

I came accross a proper analysis of this somewhere (sorry I haven't been able to dig out the ref - not very useful I know...) but, the long and the short of it was that large differences in mass are something that need to be *handled* in the sense that most solvers will throw a sicky of one sort or another unless they've been specifically designed to cope.

Do some of the commercial systems do this? - not sure but mostly not I believe, as Jon said numerical stability is already enough of a nightmare.

One approach that might be workable is to somehow "special-case" contacts on relative mass - if you imagine what happens when you throw even a fairly large thing at a tank (ie the thing bounces off and the tank doesn't really notice), you can maybe see where I'm going. I'm thinking that you might be able to treat the impact between the TV and the Sherman Tank simply as a collision between the TV and some terrain.

I'm not saying this is completely straightforward (or totally realistic) but if you have access to the solver setup code it's probably possible to slip in something of this sort just to see how it all works out.

Just an idea...



> -----Original Message-----
> From: Amund Børsand [mailto:amund at offroad.no]
> Sent: None
> To: ode at q12.org
> Subject: Re: [ODE] Large mass and accuracy
> 
> 
> I call my units "meters" and "tons". That way a car for 
> instance measures 2 wide, 5 long, and weighs 1.5. However, my 
> wishbones and rods for steering etc. are then very light, 
> maybe 1 kg = 0.001 unit, they seem to be very bouncy and 
> jerky sometimes, perhaps because of the span in numbers. 
> Also, my torques for the wheels become very small... but I 
> can't make a realistic simulation without this much span in 
> units. I also need gravity of 9.81.
> 
> I assume it all bottoms out in floating point numbers, which 
> have a certain amount of bits to represent a value, and a 
> small number of bits to place the comma. So I guess all 
> values should be centered around the same magnitude in order 
> to keep the losses/inaccuracies low, and if I'm not mistaking 
> completely it should be the same whether you center 
> everything around 1, 0.00001 or 100000. 
> 
> Come to think of it I may be ruining all the accuracy by 
> using a timestep of 0.02 (seconds), which happens at 50 fps. 
> Would I maybe be better off doing 0.2, and setting gravity to 
> 0.981, hmm, and adjusting torques and forces according to the 
> time scaling? But then I'd have to adjust mass accordingly in 
> order to avoid objects being 10 times harder to accelerate.. 
> and then the span is just as bad again, right?
> 
> I don't know what's "correct" here, maybe some of the 
> developers could enlighten us all and explain what causes 
> problems and what works, and what's the best way to do it?
> 
> ---------- Original Message ----------------------------------
> From: Michael.Rauh at Dornier.eads.net
> Date:  Tue, 15 Jun 2004 14:36:29 +0200
> 
> >
> >The examples I built so far showed that masses like 2000 kg 
> are no problem
> >if you adjust some parameters (ERP, CFM, collision params), 
> but I still had
> >problems with masses of around 50,000 kg (e.g. a large truck 
> or a tank),
> >for example the wheel joints behaved strangely. Would be 
> nice to get some
> >feedback from someone who has experience with this topic.
> >
> >Greets
> >Michael
> >
> >
> >
> >
> >                                                             
>                                                                       
> >                      Ivica Aracic                           
>                                                                       
> >                      <ivica.a at gmx.de>         An:      
> ode at q12.org                                                   
>              
> >                      Gesendet von:            Kopie:        
>                                                                       
> >                      ode-bounces at q12.         Thema:   
> [ODE] Large mass and accuracy                                 
>              
> >                      org                                    
>                                                                       
> >                                                             
>                                                                       
> >                                                             
>                                                                       
> >                      15.06.2004 13:58                       
>                                                                       
> >                                                             
>                                                                       
> >                                                             
>                                                                       
> >
> >
> >
> >
> >Hello all!
> >
> >When I ahve large mass values (e.g. 2000 for a car).
> >Could such large values cause numeric drift and instability, 
> when used
> >in an ODE simulation?
> >
> >thx!
> >macross
> >_______________________________________________
> >ODE mailing list
> >ODE at q12.org
> >http://q12.org/mailman/listinfo/ode
> >
> >
> >
> >
> >
> >
> >_______________________________________________
> >ODE mailing list
> >ODE at q12.org
> >http://q12.org/mailman/listinfo/ode
> >
> _______________________________________________
> ODE mailing list
> ODE at q12.org
> http://q12.org/mailman/listinfo/ode
> 

**********************************************************************
The information contained in this email and its attachments is
confidential.  It is intended only for the named addressees 
and may not be disclosed to anyone else without consent from
Blue 52 Games Limited. Blue 52 give no warranty that this email 
message (including any attachments to it) is free of any virus 
or other harmful matter and accepts no responsibility for any 
loss or damage resulting from the recipient receiving, opening
 or using it. 
**********************************************************************