Real-Time Physics Simulation Forum

Re: allowedCcdPenetration questions

2011-03-31T21:33:38+00:00

continuous sweeps would not detect hits unless body0 and body1 penetrate more than allowedCcdPenetration

This is the correct interpretation. I just fixed a bug, latest svn trunk should work now:
http://code.google.com/p/bullet/source/detail?r=2362

Sorry for the long delay in handling that issue,
Thanks,
Erwin

Statistics: Posted by Erwin Coumans — Thu Mar 31, 2011 9:33 pm

Re: allowedCcdPenetration questions

2011-01-27T18:22:10+00:00

I'm having exactly the same problem. There is a test at the end of btContinuousConvexCollision::calcTimeOfImpact():

Code:

if ((projectedLinearVelocity+ maxAngularProjectedVelocity)<=result.m_allowedPenetration){    return false;}

In the case of shapecasts, result.m_allowedPenetration is initialized from btDispatcherInfo::m_allowedCcdPenetration, which defaults to 0.04. Unless I'm sorely mistaken, this means that the shapecasts used in btKinematicCharacterController::stepForwardAndStrafe() will always fail given a sufficiently slow speed of movement, or a sufficiently high framerate.

I'd like to reiterate paul.dubois' question: what is the expected behaviour? Also, what is the effect of tweaking btDispatcherInfo::m_allowedCcdPenetration?

Thanks in advance,
- Steve

Edit: In case this clarification helps, here's the relevant part of the commit log from the introduction of btDispatcherInfo::m_allowedCcdPenetration (revision 1193):

erwin.coumans 08/07/2008 8:08:49 PM
+ improved KinematicCharacterController
+ improved btSubsimplexConvexCast, btContinuousConvexCollision and btGjkConvexCast to support configuration that start in touching/penetration, required for 'sliding'.

Edit 2: Ah, I see this issue has already been been reported: http://code.google.com/p/bullet/issues/detail?id=347
Almost a year old now...

Statistics: Posted by sfreeland — Thu Jan 27, 2011 6:22 pm

allowedCcdPenetration questions

2010-10-29T19:13:35+00:00

A couple questions about allowedCcdPenetration.

What is the intent of allowedCcdPenetration? Is it designed to improve gjk stability, or perhaps to allow the user to ignore grazing contacts? Or am I on the wrong track here?

Also, what is the expected effect? I previously thought it was something like a backwards collision margin -- continuous sweeps would not detect hits unless body0 and body1 penetrate more than allowedCcdPenetration. However, in this situation:

- allowedCcdPenetration is the default (0.04)
- body0 and body1 are slightly penetrating (say 0.03)
- body0 performs a short (0.03) convex sweep test into body1

I expect to receive a hit at the point where body0 penetrates 0.04 units into body1; or, perhaps an hit at t=0. Instead, it seems as if all sweeps of length < 0.04 return with no result. Specifically, the check in btContinuousConvexCollision::calcTimeOfImpact examines projected velocity instead of max penetration, which seems incorrect if there is pre-existing penetration. Is this the expected/desired behavior, or a bug?

The concrete case I'm looking at is a character proxy, which can walk substantially into collision when moving slowly.

(edit: added a bit of clarification)

Statistics: Posted by paul.dubois — Fri Oct 29, 2010 7:13 pm