Bug / Crash gravity does not apply when I manually set the fixed delta time

[lufydad]

Hello,

When I set the fixed delta time from script to 1/60 in an empty scene with just a rope, the rope just stay still, the gravity does not seem to apply on it. If I change the fixed delta time to another value the rope falls down normally.

I use Obi 6.2, Unity 2021.2.4f, and DX11.

Is there something that I can do to fixed this ?

[josemendez]

Hello,

When I set the fixed delta time from script to 1/60 in an empty scene with just a rope, the rope just stay still, the gravity does not seem to apply on it. If I change the fixed delta time to another value the rope falls down normally.

I use Obi 6.2, Unity 2021.2.4f, and DX11.

Is there something that I can do to fixed this ?

Hi!

Just set the solver's sleep threshold accordingly. This is the minimum kinetic energy above which particles are allowed to move. If you reduce the timestep, the kinetic energy during a single timestep will also be reduced, preventing any particle whose energy is below the threshold to move. See:
http://obi.virtualmethodstudio.com/manua...olver.html

You can set it to zero if you want particles to move regardless of how much energy they have. Note this may result in positional drifting/jittering.

[lufydad]

Hi!

Just set the solver's sleep threshold accordingly. This is the minimum kinetic energy above which particles are allowed to move. If you reduce the timestep, the kinetic energy during a single timestep will also be reduced, preventing any particle whose energy is below the threshold to move. See:
http://obi.virtualmethodstudio.com/manua...olver.html

You can set it to zero if you want particles to move regardless of how much energy they have. Note this may result in positional drifting/jittering.

Thank you a lot it seems to work, when I reduce the sleep threshold 
In my application the fixed delta time is entered by user, which sleep threshold should I use to be sure to not have this problem anymore and avoid drifting and jittering ? dt/10 is enough ?

[josemendez]

Thank you a lot it seems to work, when I reduce the sleep threshold 
In my application the fixed delta time is entered by user, which sleep threshold should I use to be sure to not have this problem anymore and avoid drifting and jittering ? dt/10 is enough ?

Allowing your user to set an arbitrary fixed delta time is a recipe for terrible performance, but assuming you're not concerned by this: You could use the smallest value that works with the smallest delta time you allow.

Or instead of using a very small threshold (which may not be enough for very small timesteps), you can calculate the value you need for a given timestep.
Kinetic energy is defined as 1/2 * sqr(velocity) * mass. In Obi, the threshold is mass-normalized, so not multiplied by mass.

Velocity is defined as change in position over time. In Unity, this "time" is the timestep (aka fixed delta time) value. So:

Code:

velocity = (position at the end of the timestep - position at the start of the timestep) / timestep.

Hence, kinetic energy is:

Code:

1/2 * sqr((end position - start position) / timestep).

This is the value compared to your sleep threshold to determine if a particle can move or not.
To make the threshold independent of timestep, you can do:

Code:

timestepIndependentThreshold = sqrt(threshold * 2) * timestep;

And then feed this value to Obi instead of a fixed threshold.

[lufydad]

Allowing your user to set an arbitrary fixed delta time is a recipe for terrible performance, but assuming you're not concerned by this: You could use the smallest value that works with the smallest delta time you allow.

Or instead of using a very small threshold (which may not be enough for very small timesteps), you can calculate the value you need for a given timestep.
Kinetic energy is defined as 1/2 * sqr(velocity) * mass. In Obi, the threshold is mass-normalized, so not multiplied by mass.

Velocity is defined as change in position over time. In Unity, this "time" is the timestep (aka fixed delta time) value. So:

Code:

velocity = (position at the end of the timestep - position at the start of the timestep) / timestep.

Hence, kinetic energy is:

Code:

1/2 * sqr((end position - start position) / timestep).

This is the value compared to your sleep threshold to determine if a particle can move or not.
To make the threshold independent of timestep, you can do:

Code:

timestepIndependentThreshold = sqrt(threshold * 2) * timestep;

And then feed this value to Obi instead of a fixed threshold.

Thank you for taking the time to explain.
If I may ask one last question, I don't understand the last equation of your post. Is "threshold" in the equation the member sleepThreshold of obi solver ?

[josemendez]

Thank you for taking the time to explain.
If I may ask one last question, I don't understand the last equation of your post. Is "threshold" in the equation the member sleepThreshold of obi solver ?

Hi!

"Theshold" is the (time-dependent) input threshold you want to use, this is a value you set. The equation outputs a time-independent threshold, which is the one you feed to the solver's sleepThreshold.

So the equation sits in between your hand-picked threshold value, and the one fed to Obi.

[lufydad]

Hi!

"Theshold" is the (time-dependent) input threshold you want to use, this is a value you set. The equation outputs a time-independent threshold, which is the one you feed to the solver's sleepThreshold.

So the equation sits in between your hand-picked threshold value, and the one fed to Obi.

Ok, then suppose I input 0.001 for the sleepThreshold at the beginning of the simulation and the user input 1/60 for the fixedDeltaFrame, then using your equation :

Code:

timestepIndependentThreshold = sqrt(0.001 * 2) * 1/60 = 3.33*1e-5

I should put the sleepThreshold to this value, is this correct ?

[josemendez]

Ok, then suppose I input 0.001 for the sleepThreshold at the beginning of the simulation and the user input 1/60 for the fixedDeltaFrame, then using your equation :

Code:

timestepIndependentThreshold = sqrt(0.001 * 2) * 1/60 = 3.33*1e-5

I should put the sleepThreshold to this value, is this correct ?

Yes, that should work.

[lufydad]

Yes, that should work.

Great !! Once again, thanks a lot

[josemendez]

Great !! Once again, thanks a lot

You're welcome!