Nonconvex manifolds may require additional contacts about bepuphysics2 HOT 4 CLOSED

As of the alpha, the infrastructure exists to create up to 8 contacts per nonconvex pair. The contact reduction mechanism (constrainedness heuristic) appears to work pretty well.

I did, however, arbitrarily limit the number of output contacts to 4 for now, just to see how far that can go. If no one runs into some pathological case that simply requires more contacts, this is a nice performance win.

from bepuphysics2.

After some extra work on nonconvex reduction heuristics in the beta version, it works remarkably well. Not quite ready to close the door on more contacts until more representative mesh/compound tests are done, but it's very promising.

from bepuphysics2.

A few use cases definitely do benefit from additional contacts, so it's likely that the contact cap will be raised. However, it's not good enough to just raise the cap- some cases benefit from a lower cap.

The best solution would be an improved NonconvexReduction heuristic which is able to output an appropriate number of contacts rather than just accepting all candidates up to the limit if available.

from bepuphysics2.

github, that wasn't intended as a reference shoo

also, the nonconvex reduction heuristic was improved; shipping with 4 contact cap. Seems fine until proven otherwise.

from bepuphysics2.