RBD: spheres inside a thin hollow box

[guache]

I don't have experience with sims, but I need to set up a simple RBD sim. I have 15 balls inside a thin, hollow box (a bit thicker than the balls). I want the balls to fall and collect at the bottom of the box. The balls fall down, but the box doesn't hold them in. They collide with features of the box, but, in the end they fall out of the box, as if the box's walls had "holes". The box is a Static Object (Use Surface Collision), its poly normals point to the center, as you'd expect in a hollow object. The balls are an RBD Point Object (Use Volume Based Collision on). I use RBD Solver, I tried Bullet (same setup), but the balls just fall straight down. Questions:

1) For a hollow box, is having just the inner walls (with normals pointing inward) OK, or do the walls need "thickness", like a shell?

2) Why is the box detected for collisions by the basic RBD Solver (not very well, but still), while in Bullet the balls fall right through the box?

 

[underscoreus]

1. Yes, a shell is definitely preferable especially when you want stuff to collide with the "insides" of an object like this. In fact, if you can, make the thickness as high as you can to help the sim out as much as possible.

2. The reason for the collisions being different might come down to the default way each solver creates the collision geometry. If you go to the static object node and turn off the "Display geometry" check box and then, in the "Collision" tab enable the "Show collision geometry" checkbox you should be able to see what the solver is actually working with/what the solver is actually colliding with. By default, it's probably not a perfect representation of your geometry just because they are trying to optimize the geometry to solve faster. For bullet the default is a convex shape, meaning that in simple terms it only supports faces that curve outwards, never inwards. For a shell, like yours you'd probably have to select "Concave" or set up your collision geometry in sops using the "convex decompose" node or something similar. For the RDB solvers collision, you might just have to increase the uniform samples, but I'm less familiar with it since I mostly prefer to use the bullet solver.

Hope that info can be useful.

[guache]

Thanks. I'll create a thick shell and see how it goes. For Bullet, I'll try to re-do the box by treating the individual walls as separate objects (making sure they're thick and convex)

Edited April 25, 2021 by guache