Capsule vs Capsule Collision Make sure the penetration depth is not zero

[robertocapuano]

Sometimes assertion fails in NarrowPhaseInfoBatch::addContactPoint():
assert(penDepth > decimal(0.0));
This happens for Capsule/Capsule collision check.

Probable cause is float approssimation in
bool CapsuleVsCapsuleAlgorithm::testCollision(NarrowPhaseInfoBatch& narrowPhaseInfoBatch, uint32 batchStartIndex, uint32 batchNbItems, MemoryAllocator& /*memoryAllocator*/) ;

I inserted a double check on penetrationDepth.

if (penetrationDepth > 0)

The logic is the same of SphereVsSphereAlgorithm::testCollision().

[AthosArantes]

Is the sqrt() really needed? Why not compare the difference of closestPointsDistanceSquare and sumRadiusSquare with an epsilon instead?

[robertocapuano]

I tried with
if ( fabsf(closestPointsDistanceSquare - sumRadius * sumRadius )<eps )

with different values for eps:

• MACHINE_EPSILON
• 1e-7

but after few seconds of simulation I always get
assert(penDepth > decimal(0.0));
in NarrowPhaseInfoBatch::addContactPoint

[AthosArantes]

After reading a second time, what about something like
std::max(sumRadius - closestPointsDistance, MACHINE_EPSILON)
in

reactphysics3d/src/collision/narrowphase/CapsuleVsCapsuleAlgorithm.cpp

Line 192 in c2201c5

decimal penetrationDepth = sumRadius - closestPointsDistance;

[robertocapuano]

It works well, I tried it for a few minutes without errors.

[robertocapuano]

Reverted the previous change and pushed the fix on this branch.

[DanielChappuis]

Thanks a lot for your pull request.
I have merged your first commit into the develop branch and refactored it a bit myself before your last two commits.
Are you able to try the develop branch with your test and see if the issue is now fixed?

[robertocapuano]

Just tried it for a few minutes: it works. Thank you.

[DanielChappuis]

Just tried it for a few minutes: it works. Thank you.

Thanks a lot for your feedback. That's very helpful.