Real-Time Physics Simulation Forum

Computing distance

2006-04-05T22:09:48+00:00

Thanks for the quick response!
It's really hard to work on the dynamics part alone (without knowing about stuff in CD).

Statistics: Posted by ngbinh — Wed Apr 05, 2006 10:09 pm

Computing distance

2006-04-05T21:46:16+00:00

This is called contact manifold generation. There have been a couple of similar postings/questions recently.
When using GJK, you need to gather multiple points. Several methods exist, Bullet uses currently Persistent Manifold that adds one closest point at a time, and keeps a small point set persistent over frames. Please see other postings about this:

http://continuousphysics.com/Bullet/php ... .php?t=288

Another approach, very suitable for polyhedra and in combination with Separating Axis Test based collision detection: you can explicitly build the manifold by finding two parallel planes and perform clipping using edge-planes.
If I got time I might start writing some white-paper about Contact Manifold generation in combination with GJK.

Statistics: Posted by Erwin Coumans — Wed Apr 05, 2006 9:46 pm

Computing distance

2006-04-05T21:25:10+00:00

Hi!
Another question:

Let me describe the scenario first:
Two body A,B are close (or already) in contact, I'd like to get all the features distance that < some epsilon not just the closet one?

That features of collision detection are essential for my current time stepper to prevent intepenetrations, not just recover from penetration.

A side question: Anyone know of any physics engine implementation using that technique?

Thanks

Statistics: Posted by ngbinh — Wed Apr 05, 2006 9:25 pm

Re: Computing distance

2006-03-19T18:19:23+00:00

Hi!
Anyone know of algorithm that compute the distances between two bodies as well as the penetration depth?
Could I modify Bullet to get the distances?

I need that distance because some time steppers use "prevent penetration" not "penetration correction" approach, so the distances are needed.

I'd guess SWIFT++ could return distance but no penetration
thanks

SOLID 3.5 computes both the distance and the penetration depth for any combination of convex shape types. SOLID 3.5 can be downloaded from http://www.dtecta.com.

Cheers,

Gino

Statistics: Posted by gino — Sun Mar 19, 2006 6:19 pm

Computing distance

2006-03-17T23:21:27+00:00

OK. That's great!
I didn't really dwell into Bullet, just messing around with higher level interfaces.

Thanks alot.

Statistics: Posted by ngbinh — Fri Mar 17, 2006 11:21 pm

Re: Computing distance

2006-03-17T22:40:16+00:00

Hi!
Anyone know of algorithm that compute the distances between two bodies as well as the penetration depth?
Could I modify Bullet to get the distances?

I need that distance because some time steppers use "prevent penetration" not "penetration correction" approach, so the distances are needed.

I'd guess SWIFT++ could return distance but no penetration
thanks

Bullet GJK gives you already the closest distance. See in ConvexHullDistance demo:

Code:

PointCollector gjkOutput;GjkPairDetector::ClosestPointInput input;input.m_transformA = tr[0];input.m_transformB = tr[1];convexConvex.GetClosestPoints(input ,gjkOutput);

Code:

struct PointCollector : public DiscreteCollisionDetectorInterface::Result{SimdVector3 m_normalOnBInWorld;SimdVector3 m_pointInWorld;SimdScalarm_distance;//negative means penetration...

There is currently penetration depth estimation, and a EPA implementation is contributed but not added to CVS yet. And more high-level interfaces, but in those cases there is broadphase culling, which prevents calculation of closest distance if objects are too far apart.

Erwin

Statistics: Posted by Erwin Coumans — Fri Mar 17, 2006 10:40 pm

Computing distance

2006-03-17T21:40:54+00:00

Hi!
Anyone know of algorithm that compute the distances between two bodies as well as the penetration depth?
Could I modify Bullet to get the distances?

I need that distance because some time steppers use "prevent penetration" not "penetration correction" approach, so the distances are needed.

I'd guess SWIFT++ could return distance but no penetration
thanks

Statistics: Posted by ngbinh — Fri Mar 17, 2006 9:40 pm