CCD reports a contact point far from the shape about physx HOT 4 CLOSED

Unfortunately huge edge length differences (sliver triangles) are notoriously problematic especially with single precision. If you can avoid such triangles by e. g. using a different tessellation, that would be great. Debugging is difficult for us unless you could provide the full code to reproduce the described issue.

I fixed it by checking wheather Q0 == Q1 .

Indeed we need to avoid such annoying triangles, thanks for your advice!

from physx.

Thanks for the report, @pkusilence - we'll have a look. Internal tracking: PX-4666.

from physx.

It turns out that the original tiangle has a huge edge length ratio, i.e La = Lb >>>Lc. Due to the float point differences between platforms, the support function found different vertices if the search direction is nearly perpendicular to La /Lb。

from physx.

Unfortunately huge edge length differences (sliver triangles) are notoriously problematic especially with single precision. If you can avoid such triangles by e. g. using a different tessellation, that would be great. Debugging is difficult for us unless you could provide the full code to reproduce the described issue.

from physx.