Custom Damping Forces

[cgpauli]

Hi,

i am trying to use a custom force callback to generate angular and linear damping forces. This is what the callback looks like: (i am using OgreNewt)

Code: Select all

Ogre::Vector3 angDamping;
angDamping = -10*body->getOmega();
body->addTorque(angDamping);

Ogre::Vector3 damping;
damping = (-3)*body->getVelocity();
body->addForce( damping );

Thereby i am experiencing the following issues: First the body behaves like expected, slowing down continously, but when it reaches an omega of 0.0607868 respectively a damping torque of -0.607868 (only having a rotation about one axis at the start), it suddenly stops to omega=zero, which looks very bad! Even more annoying, if i disable auto-sleep for the body, it keeps revolving at an omega of 0.0607868 till the end of time. If i set the standart newton parameters for damping instead, i do not experience these issues! Is there any way to fix this respectively am i doing something wrong? Thanks for your help !

btw: i use a physics update rate of 100Hz

[Julio Jerez]

what it it stops or keep moving? I am not clear what you mean

[cgpauli]

what it it stops or keep moving? I am not clear what you mean

well, there are 2 different cases:

- if i enable auto-sleep: The body suddenly stops to rotate when omega gets smaller than about 0.0607868 rad/s.

- if i disable auto-sleep: The body keeps revolving at a constant omega of about 0.0607868 rad/s as soon as the body has slowed down to that value of omega .

Thank you for your help!

[Julio Jerez]

if you disable autosleep the body will rotate controll by the net torque.
set all drga to zero, and you are in total controll of the body.
There is nothing in the engine othe than the coefficent of linai and angular damping that act over the body.

I will have to see a demo of that because I do no thing that happens

[JernejL]

Change angular dampening to zero, see how it behaves then.

[cgpauli]

Hi,

thanks for the answers! I have the standart damping forces/torques set to zero, so that's should not be the problem. Actually, if i use the standart damping forces/torques instead of my custom ones it works all well...
@Julio: What do you mean with "demo", the code of the complete program? Sorry if this is a stupid question !

[Julio Jerez]

I mean a test program that exposes the problem and that I can attach to the newton DLL from VS, so that I can check in debug mode.

[cgpauli]

Ok, here is a link to the program that exposes the problem: http://www.megaupload.com/?d=PX6AKKT7

It would be very kind of you if you could have a look at the problem.
Thanks in advance!

p.s: the program shows the model of a submarine, you can apply local forces/torques via the arrow keys, otherwise only the custom damping forces mentioned above are applied to slow the motion down. You will see that the model does not slow down to zero. ( auto sleep is disabled )

[cgpauli]

@Julio: If you don't have the time to look at my problem ( which would be very understandable ) or if there is a something wrong with the test app, could you please tell me, because in the first case i would have to switch to another physics engine, since i seem to be unable to find an error in my own code and the current behaviour is not acceptable for my purposes. Thank You!

Of course, if any one else has thoughts about this, i would appreciate it very much.

[Julio Jerez]

Sometime I forget to test problems people post because I cannot run then during the week.
And then when the weekend comes I forget where the post was, people usually send me a PM when they post a test or test.

Ok I run this and you are right there is a bug in the integration when the damp is set to zero
and the force was lower that the minimum threshold. The solver was stopping applying the force as if the internal damp was still on,
but the damp was zero which lead tp the body moving a constant speed.
I fixed it, the force do not have to check if the damp is off, it needs to be applied all the time.

That was a big bug if people set damping to zero, I do not kwno why it took so long to come up.
Thanks for the report. Get 2.24 here
http://www.newtondynamics.com/downloads/NewtonWin-2.24.rar

I am not ready to post it to the download yet since I am changing the editor GUI, but the fix is there, just get the dll.

[cgpauli]

Wow, great service, thanks a lot my friend !

[cgpauli]

Hi,

there still seems to be some kind of force or velocity threshold although autosleep is disabled. If velocity or angular velocity of the body becomes smaller than about 0.02, then the body freezes appruptly. To get it moving again, you have to apply a force/moment grater than some minimum threshold. Would be great if you could fix this!

Test App (Newton 2.24):

http://www.megaupload.com/?d=90IA59HG
(you can apply forces/torques via arrow keys)

[Julio Jerez]

did not I fixed this bug in 2.24?
This is the same submarine demo you sent before.

this seems to be the same bug as this othe thread
http://newtondynamics.com/forum/viewtopic.php?f=9&t=6173&p=44456#p44456

I will check again but I believe I fixed this, how is this different from what I fixed

[misho]

Hi,

there still seems to be some kind of force or velocity threshold although autosleep is disabled. If velocity or angular velocity of the body becomes smaller than about 0.02, then the body freezes appruptly. To get it moving again, you have to apply a force/moment grater than some minimum threshold. Would be great if you could fix this!

Test App (Newton 2.24):

http://www.megaupload.com/?d=90IA59HG
(you can apply forces/torques via arrow keys)

I can confirm that, the same thing is happening to me at around 0.03 RPS. This bug is already logged at http://newtondynamics.com/forum/viewtopic.php?f=9&t=6173&p=44456#p44456, as Julio says.

Misho

[Julio Jerez]

Ok I was looking at this and I see what you say
I run teh demo slowely bachwar and teh velocity start to sloa daown gradually and as soon as it reach
veloc {m_x=-0.031622492 m_y=0.00000000 m_z=0.00000000 ...}

It is damp very rapidly to zero.
that is because ther have a cap velocity of abs (0.03) which I consider very smal whent doing 32 bit float and

Here we need to deal with the reality fo Float arithmetic on a computer,
I cannot let the numbers run wild and become very low or very big because then I run into floating point degeneracy which slow down the float unit.
I reduce the attenuation by a factor of 1000, but I can not make it zero, that will be very disatrous for the engine.
Not for you by for many users that let body update perpetually and tehn as they slow down teh floting point using also slow down nomatlizing denormal numbers intenally.
those are the problem we have to deal with 32 and 64 bit floats, and we most accept that fact.

even if the calculation was done in double presition you cannot really let a values be infinitelly small or infinitleey large you must use error bound arithmetic or
Adaptive precition aritmetic using predicates which mean that a value will be considered zero based of the ration betwenn teh magnitud on the larger value
involve in the calculation to teh machine presition (this is quite conplex for a realtime physic library bu it is use in soem intenal algorithms in newton)

Anyway with the tolerance set to 0.9999 intead 0 0.99 the velocity reach zero when it becomes smaller than 0.03 but it take a lot longer to do so.
in theory the velocity should reach zero in infinite amount of time, but like I said before this is quite desatrus for the engine.