Real-Time Physics Simulation Forum

I seem to be running into the issue that was discussed in this thread. The op indicates that the proposed solution of generating an aabb around the penetration vectors and then using the corner that is furthest from the origin of the aabb "works". However, I'm confused as to what the corner vector of the aabb is used for? Is it the distance? If so then how is the direction calculated? Is it the direction? If so then how is the distance calculated? Is it both? If so then what happens when multiple corners of the aabb are the same distance apart?

EDIT: Also, after thinking about this further...if you're using the origin of the aabb, wouldn't the eight corners always be equidistant?

OK, so I misunderstood what was meant by "origin" in that original post. I was thinking they were referring to the origin of the AABB, but they're referring to the origin of the world. Generating an AABB around your penetration vectors and finding the corner with the greatest length then yields the vector to be used for your direction and distance. Below are some screenshots to help visualize the concept further: