A few questions about iterative solver...
Posted: Tue Jul 26, 2005 7:11 pm
Hi,
Good to see a forum dedicated to physics and nice work on the Bullet project!
As an introduction (and a shameless plug) I'll post here what I've done so far on the subject.
I do post my progress on Gamedev quite frequently so it might not be new stuff: -Physics WIP-
It's impulse based and pretty much a hack since static contacts are modeled using point/point constraint in order to minimize the drift produced by the sequential processing of the contact points. In order to make it less of a hack I'm planning to try solving resting contacts using a global method but...
1) Do iterative solvers eliminate all drift and jittering or do you still need to stabilise the results in some ways?
2) Are they significantly better at handling large mass ratios?
My primary concern is performance. My current exe do brute-force contact caching (my todo list is too long ) so it get very slow when the contact count rises but otherwise it can be made pretty fast. Could I see much performance/stability gains with a global solver?
Oh and... just in case, this is all a hobby and my Math skills are pretty weak so, unless you need to, please don't go too hard on the Mathematic terms .
Thanks!
Good to see a forum dedicated to physics and nice work on the Bullet project!
As an introduction (and a shameless plug) I'll post here what I've done so far on the subject.
I do post my progress on Gamedev quite frequently so it might not be new stuff: -Physics WIP-
It's impulse based and pretty much a hack since static contacts are modeled using point/point constraint in order to minimize the drift produced by the sequential processing of the contact points. In order to make it less of a hack I'm planning to try solving resting contacts using a global method but...
1) Do iterative solvers eliminate all drift and jittering or do you still need to stabilise the results in some ways?
2) Are they significantly better at handling large mass ratios?
My primary concern is performance. My current exe do brute-force contact caching (my todo list is too long ) so it get very slow when the contact count rises but otherwise it can be made pretty fast. Could I see much performance/stability gains with a global solver?
Oh and... just in case, this is all a hobby and my Math skills are pretty weak so, unless you need to, please don't go too hard on the Mathematic terms .
Thanks!