Hi,

In my Physics Engine I use Polyhedra for my objects & I get the Intersection Polyhedra & take the Centre of Mass as the Collision Points.

Sometimes creating the Intersection Polyhedra fails due to floating-point rounding errors

Are there more reliable ways of creating Collision Points?

cheers

I gave two talks on this topic:

"Robust Contact Creation for Physics Simulations." 

http://media.steampowered.com/apps/valve/2015/DirkGregorius_Contacts.pdf

"The Separating Axis Test Between Convex Polyhedra." 

http://media.steampowered.com/apps/valve/2013/DGregorius_GDC2013.zip

Thanks Dirk, creating manifolds seems to be an entire subject on it's own

1 hour ago, Cacks said:

Thanks Dirk, creating manifolds seems to be an entire subject on it's own

Yep. 90% of groundwork to get a physics engine running is dealing with collision detection.

@Dirk Gregorius,

I had a quick read of your presentations.

You use the polyhedra overlap vertices to create manifolds

How would you deal with 2 long perpendicular polyhedra cylinders which collide?

No vertices would overlap...

Exactly the same. The algorithm will handle this. Just think of two 2D boxes which build a cross. I recommend going through this example and draw each step on paper. It is a very simple case. Nothing special about it.

@Dirk Gregorius,

yes I see now, even though the vertices are not inside SAT will show an overlap!

Thanks

@Dirk Gregorius,

I am having trouble with the 'QueryEdgeDirection' algorithm in your 'Robust Contact Creation for Physics Simulation' presentation

I understand I want to find the 2 edges with least penetration, but I don't understand how this algorithm can find them?

I have gotten 'QueryFaceDirection' to work well, thanks

Did you see the sample code? In a nutshell, just test each edge combination. The Gauss-Map is just an optimization to prune edge pairs quickly. If you have a more specific question what you don't understand I might be able to help!

I can see the edge algorithm is very much like the face algorithm.

The Face algorithm finds the largest separation and the edge algorithm seems to do the same.

I have a few problems with the example I have attached.

Polyhedron P1 is rotated 45 degrees around the Y axis. I get the cross product between edges E1 & E2: this gives either the +X or -X axis.

Problem 1: I can't determine the direction of the axis because the dot product will always give zero

Problem 2: If I can determine the direction of the X-axis there is a separation between E1 & E2 & the algorithm will fail

Maybe I have misunderstood?

Thanks