GImpact in Bullet 2.73 "freezes" simulation

[faichele]

Hello!

I hope I've chosen the right subforum for my questions.

Using Bullet 2.73 with GImpact as collision algorithm, I run into an issue that I can't resolve:
I have a static box acting as ground plane and a rigid body using a concave collision geometry (see the attached screenshots); if I position the body about 30 "units" above the static box and just let it fall under the influence of gravity (-9,81 kg/ms^2 along the Z-axis), the simulation "freezes" after about 400 iterations of the Bullet engine; the application itself remains responsive (I use OpenSceneGraph for visualization), i. e. I can still move the camera around, but the rigid body seems to come to a "final rest", even if it is in the process of toppling over after hitting the ground. I have stepped into my simulation loop with a debugger to verify that the "stepSimulation" calls to Bullet still occur, and they definitely do.

I tried the same free-fall experiment with a simple cube from different heights, and these simulation runs didn't "freeze", but kept running until the cube came to a rest (I rotated the cube initially, so that it would bounce off the ground a few times after hitting the ground).

Most likely the issue is me overlooking some important parameter setting or using Bullet incorrectly in some other way; any suggestions and tips are highly welcome.

Thank you!

Fabian Aichele

screenshot_1.jpg

screenshot_2.jpg

[ola]

Hi, you can try to disable deactivation of rigid bodies:

b->setActivationState(DISABLE_DEACTIVATION);

Or you can modify the threshold velocities for when bullet does deactivate a rigid body. The default values are a bit "aggressive", in order to deactivate as much as possible and save CPU.

from btRigidBody.h:
void setSleepingThresholds(btScalar linear,btScalar angular)
{
m_linearSleepingThreshold = linear;
m_angularSleepingThreshold = angular;
}

Hope that helps,
Ola

[faichele]

Hello!

Thank you very much, that did the trick! I'll stick with completely disabling deactivation for the time being.

With best regards,
Fabian Aichele

Hi, you can try to disable deactivation of rigid bodies:

b->setActivationState(DISABLE_DEACTIVATION);

Or you can modify the threshold velocities for when bullet does deactivate a rigid body. The default values are a bit "aggressive", in order to deactivate as much as possible and save CPU.

from btRigidBody.h:
void setSleepingThresholds(btScalar linear,btScalar angular)
{
m_linearSleepingThreshold = linear;
m_angularSleepingThreshold = angular;
}

Hope that helps,
Ola

[sparkprime]

I wonder if sleeping should be disabled by default since it's only a performance feature and seems to confuse a lot of new users.

[faichele]

Hello!

Sleeping being turned on by default will have its reason, thinking of the time-frames one usually has available for physics processing in an interactive real-time application like a modern 3D computer game.
Remembering a presentation by a developer from Rockstar Games, about 25 - 30ms per game loop iteration = physics + graphics + AI. You need every possible optimization you can get there, as far as I can imagine.

[sparkprime]

New users don't need performance, they need a simulation that works without having to be configured for their situation. Performance tuning always comes after getting things to work.