How to deal with getting stuck in sequential impulse?

Started by
17 comments, last by Finalspace 6 years, 10 months ago
So i am back:

The implementation works beatifully, i get perfect line segments without any internal edges due to connected edges, but unfortunatly the internal edge issues are still present for vertex vs vertex contacts, when trying to fall off - see the following video demonstrating the progress and the issue:




Should i just drop any vertex vs vertex contact entirely - detected by some distance threshold + dot product???
Advertisement

Oh definitely drop vertex to vertex cases. These can be handled implicitly by face to face! I imagine if you drop vertex to vertex your solution will work pretty well.

Depends on what you want. With connected line segments the normal would be between the angle the two segments form, so something closer to the locked one is probably what you are after.

I solved it by just porting over the contact generation from box2d directly. At start i dont wanted to do that, but after looking into the box2d collision source very briefly i decided to port just the collision stuff over, because its really well implemented in a efficient way. I would have ended up remembering the adjacent edge informations anyway - which solves the getting stuck problem in a simple way.

Also i think storing and computing all manifolds in local space is really smart. I always stored and computed in world space which is harder and bad for contact caching coherence.

Right now this is all i need, so i am happy and can finally continue on my actual game code.

But when i need more than this, i will just use box2d directly and use a custom memory allocator.

This topic is closed to new replies.

Advertisement