Warm-starting for 3D rigid bodies

I don't compute any IDs for contacts and instead use the previous contact point in the local space of each object to identify matches. If the euclidean distance between the old and new points is less than a threshold then I consider them to be the same. This fixes your issue from (1). The concept of a contact ID doesn't make very much sense with quadric shapes (spheres, capsules, cylinders).

For #2, I can see how keeping invalid contacts around can help with stacking stability, but be aware that that might cause you to get wrong results in other circumstances. Your physics simulation will no longer be as accurate if objects are not statically stacked. I wouldn't keep invalid points around for more than 1 or 2 extra frames. I think I tried this once to deal with jitter but ultimately removed it.

Thanks Aressera, that makes sense. For the distance check, do you check that both of the contact objects local space distance is more than your threshold then you don't apply the impulse, or just if a single objects local space distance has change. I'm thinking of the scenario where we have a cube sliding down a ramp, the cube's local coordinates for the contact are the same frame-to-frame, but the ramps changes as the cube slides down. If the cube slides fast enough it then the distance check from the ramps local space would fail.

You are correct, a match is found if either distance in local space is less than the threshold, to handle the exact case you describe.

Thanks so much Aressera!

Hi, I'm sorry for entering this conversation, but I don't want to open another thread.

I'm trying to compute the contact points using the method Aressera kindly described, but in my tests GJK-SAT/EPA always finds the same point in local space (+ or - some threshold). How do you force it to compute a different contact point (in 3d)?

Thank you in advance.

This might be helpful. I go over a lot of different shapes and different approaches

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

fantasyAlgo said:
I'm trying to compute the contact points using the method Aressera kindly described, but in my tests GJK-SAT/EPA always finds the same point in local space (+ or - some threshold). How do you force it to compute a different contact point (in 3d)?

That's expected until the objects move as a result of the first collision point. GJK/EPA only find the point with greatest penetration. You start with just 1 contact point, then that point causes a torque on the object(s) which causes a different point to be found on later frames. Use a cache of contacts (maximum 4 typically) to make a stable manifold. If you want to do better you can do a more complicated calculation like Dirk describes to get a full contact manifold on the first frame of the collision.