BulletSolver Problem

[patis]

Hello, I have noob question.

I'm animating the sphere collision by keyframe.

But, I get some errors and the collision goes through the active object.

Particularly, "Target geometry does not contain a name point attribute" error message is strange...

Is there anything way for fix to these...?

Thank you

 

Destruct Bullet.hipnc

Edited November 30, 2020 by patis

[Noobini]

think you're missing an assemble>create Packed geo before feeding into rbdconstraintproperties

[patis]

Thank you sir!! Add assemble, BulletSolver was worked 

But I still have error message. What is this...

 

Destruct Bullet v2.hipnc

Edited November 30, 2020 by patis

[nuki]

hey! the assemble creates a 'name' attribute unique to every piece. this attribute needs to be present on constraints too, else there would be no info which piece is constrained to which. also, the constraintnetwork was not set up to fetch any geometry, so no constraints were created at all.

the collision sphere also throws an error. I think the rbdpack sop is designed to keep the network clean when dealing with the bullet sop solver. so you can deal with a single stream instead of three and separate them again when needed with the rbd_type attribute / rbdunpack. Consider it another layer of packing, meant to keep components (geo, cons, proxy) of the same object together for convenience. It does not set up a "name" point attribute, which is required for the packed bullet workflow. Keep in mind that you can pack geo into packed geo as often as you'd like, you'll just have to process the right 'level' in the end. its actually pretty useful (hence the 'iterations' parameter on the unpack sop) - you could pack pieces into walls, walls into buildings, buildings into cities...then unpack it down to the pieces again before simming. the name attribute has to be unique tho.

for convexdecomposition to do anything you'll have to set the max concavity < 1, or it will be the the same as a shrinkwrap.

in terms of performance the order of operations can be crucial. you put the convexdecomp after the point deform and the transform animation. Hence it will be evaluated every timestep, as the upstream geometry changes. If you evaluate it before all of that it has to be cached only once. Btw, that pointdeform doesnt do anything meaningful and instead of animating the object, then freezing it with timeshift you could just get the static geo from further upstream

TL:DR - ''Packed' does not inherently mean your geo is ready to be used with bullet.

 

 

DestructBulletv3.hipnc

Edited November 30, 2020 by nuki

[patis]

Thank you sir! And Thank you for your very detailed advice!!! My knowledge has been very organized...

Also, fix to ground tip was so useful. I learned a lot of technique..!