Real-Time Physics Simulation Forum

tall cylinders in ccdphysics demo

2006-10-13T21:25:57+00:00

I just checked this long-cylinder case, and the effect is easy to replicate. I compared it by running the ODE quickstep solver, and it behaves the same, if you disable the 'rotational force'. But as Dirk mentioned, adding the velocity-dependent/gyroscopic/rotational force helps a bit, but adds instability.

Increasing the friction seems to be one workaround to reduce the effect, because it reduces the angular velocity naturally. Perhaps some damping on the angular velocity might help too.

Thanks for pointing this out,
Erwin

PS: In the next Bullet release you can toggle the Bullet SequentialImpulse constraint solver with ODE quickstep solver to compare results. During the recent refactoring this comparison functionality got lost.

Hi,
i have added a 5th shape to the array of shapes in the appccdphysicsdemo which is a cylinder 1x7x1 in size and replaced the brick wall with an x row of this new shape. when they fall over they do not seem to be correct... whatever 'topple over' angle they are at if they are spinning on their base they do not fall for ages they simply roam around the world. i have been told this is not happening but i swear it is. i have wasted ages on this problem and it is driving me mad - could someone please make the simple change and test on their system and see if it looks correct to them.

Statistics: Posted by Erwin Coumans — Fri Oct 13, 2006 9:25 pm

tall cylinders in ccdphysics demo

2006-10-13T12:19:52+00:00

This is because Bullet ignores the gyroscopic (like most engines) term for the update of the velocity. Simply add this to your external force and you will be fine. Note that this term on the other hand can lead to instabilities and make the system explode...

Statistics: Posted by Dirk Gregorius — Fri Oct 13, 2006 12:19 pm

tall cylinders in ccdphysics demo

2006-10-13T09:35:47+00:00

Hi,
i have added a 5th shape to the array of shapes in the appccdphysicsdemo which is a cylinder 1x7x1 in size and replaced the brick wall with an x row of this new shape. when they fall over they do not seem to be correct... whatever 'topple over' angle they are at if they are spinning on their base they do not fall for ages they simply roam around the world. i have been told this is not happening but i swear it is. i have wasted ages on this problem and it is driving me mad - could someone please make the simple change and test on their system and see if it looks correct to them.
Regards
PP

Statistics: Posted by Proctoid — Fri Oct 13, 2006 9:35 am