Is featherstone suitable for a simple structural analysis?

Post Reply
AlanQ
Posts: 2
Joined: Sat Sep 24, 2016 6:01 pm

Is featherstone suitable for a simple structural analysis?

Post by AlanQ »

Hello there.

I'm making an educational game where you build 3D structures but they have to be solid and obey realistic physics. I'm using at the moment a finite element solver but it's not as fast as I'd like to (stiffness matrix generation is slow) and also it doesn't support some element types I'd like to have.

I'm looking for alternatives to that, I don't need something 100% correct, but I need something that's consistent across different hardware, I've tried Physx but besides being very fast it's not really suitable because you don't have the more stable solvers like in Bullet (NNCG/MB) and even when I increase the sim quality to the highest level possible for my specs requirements it still produces extremely different results when using joints, presumably because floating point errors add in a kind of chain reaction of inaccuracy. Even on the same hardware a simple tower will be unstable/stable only by 'moving' it a little bit around.

I don't need perfect determinism but I want players to share their creations and they should work somewhat similarly for most cases. It's OK to fail but only for some cases. I'll notify the users if the results aren't expected to match the original because of (sw/hw) architecture differences.

[s]So, I'm wondering if the new multibody feature is suitable for that. There's an example on github that shows how to get data from constraints so I think it could be used for 'breaking' them, but I'm somewhat lost on how to actually deal with multibodies... do you have to create them like a compound multibody object beforehand (like in the examples, they're so complicated), can't you just add ordinary rigidbodies/colliders/joints around like usual and tell bullet to magically resolve all of that using featherstone?[/s]

If MB is fully independent from the standard constraints, are similar ones available for all of them or are you limited to a subset? I'm really struggling to make sense of multibody, it seems so much complicated to create the stuff. If you guys could point me to a tutorial or explain those points clearly I'd appreciate, or if Bullet (even using feathrestone) just isn't suitable for that application then perhaps you guys know some fast LGPL/MIT licensed lib I could use for that purpose. The one I'm using atm is pretty fast (BriefFE) but it doesn't even support cables and such. It's for a game so speed is more important than accuracy. There are some alternative 3D truss solvers but they're either too slow or just too limited (e.g.: elements necessarily need to have all ends connected, etc.).

On a side note, the captcha to register here is crazy stuff imho... I had to Google the book title (I thought it was 10/11, 7 or 16 at first because of Animal Farm) and that random vector math also is not very human friendly imho.

Thank you.
Last edited by AlanQ on Sun Sep 25, 2016 4:25 pm, edited 1 time in total.
AlanQ
Posts: 2
Joined: Sat Sep 24, 2016 6:01 pm

Re: Is featherstone suitable for a simple structural analysi

Post by AlanQ »

OK, after some research it seems featherweight currently is a no-go for that purpose... but what about MLCP? Maybe it could be an option but what scares me is that I read both Dantzig and Lemke may 'fail'. I don't know what would cause that and the implications. How could I check if the simulation is converging to a stable solution while it's substepping?

On a related question, I'm not planning to use collisions from Bullet and I see that you can't have a rigidbody without collider although there's an empty collider class. If I'm never going to have colliders, should I care about that, I mean, the broadphase routine I guess will still loop all colliders and I wonder if the difference between having empty colliders and somehow suppressing the broadphase is negligible. I guess it is but you never know.
User avatar
Erwin Coumans
Site Admin
Posts: 4221
Joined: Sun Jun 26, 2005 6:43 pm
Location: California, USA
Contact:

Re: Is featherstone suitable for a simple structural analysi

Post by Erwin Coumans »

AlanQ wrote:OK, after some research it seems Featherstone (spell-corrected) currently is a no-go for that purpose.
Why not? It seems btMultiBody (Featherstone based) is rigid, provides you with joint reaction forces etc.
Post Reply