Unstable "Kapla" Box Primitives (Dynamica Problem)

[RBD]

I'm trying to simulate stacked "Kapla" blocks using box primitives but they start jumping as soon as a few are stacked. Kapla planks have a size ratio of 1:3:15. Is the problem due to the lenght? Will this idea not work? Tried increasing substeps... didn't help.

Here's an example video of the situation (floor is a plane primitive, blocks are box primitives, one stack with blocks sized 1:3:15 units and one stack with blocks 1/2 size of that; tried larger... didn't help) using MayaPlugin built from Bullet rev 1765 (also tried 1751):
Kapla Unstable

Thanks.

[gjaegy]

Not sure at all, but have you tried to enable "split impulse" (you need 2.75 for that) ?

[RBD]

Thanks, but that doesn't appear to help. I can't get these stable unless I stack them at their center of gravity. I guess it is due to the longer they are, the more even small variations in angular momentum will wreak havok.

I'll have to switch to Jenga type blocks, they're more compact and therefore more stable. But still, if I stack'em high (25 or more), then the stack will bob up and down as the forces propagate through it. I wonder if there is any way to eliminate bobing for tall stacks of rigid bodies if real-time performance is not required? (was using fixedTimeStep = 1/120 & maxSubSteps = 1000; tried fixedTimeStep of 1/240 & maxSubSteps 10000, bob becomes a fast vibrate, 25+ stack still not stable).

[Erwin Coumans]

It looks like the collision detection goes wrong in the initial video, are those btBoxShape objects? Are you setting the collision margin to zero?

Can you share the Maya file, attached to this forum as a zip file?
Thanks,
Erwin

[RBD]

Yes, shapes are btBoxShape and collision margin appears to be the default. Essentially, I'm using the unchanged source code from r1765, only tried tweaking the Dynamica stepSimulation params later to test.

To avoid problems with different Maya version scene files, I've attached a MEL script. With the Dynamica plugin loaded in Maya, in a New Scene, source the MEL script from the Script Editor. Set your timeline playback range end time high (for real-time playback, right-click the timeline, set Playback Speed to Play every frame, Max Real-time).

Thanks!

[Erwin Coumans]

How did you tweak the stepSimulation?

It is fine to use something like

Code: Select all

world->stepSimulation(deltaTime,10,1./240.);

Does the mel script generate the scene from the movie? If not, please attach a zipped Maya .mb or .ma file.
Thanks,
Erwin

[RBD]

How did you tweak the stepSimulation?

It is fine to use something like

Code: Select all

world->stepSimulation(deltaTime,10,1./240.);

Thanks, I tried

Code: Select all

m_dynamicsWorld->stepSimulation(dt, 10000, 1.0 / 240.0);

It seemed to work well, but tall structures vibrate instead of bounce (fast bounce I guess).

Does the mel script generate the scene from the movie?

Yes, it's the same structure as in the movie; entire scene in movie was created using this script only, nothing else, except movie shows two structures with different height and Kapal block sizes (configurable in script: $height and $blocksize). I get the same result with one or more structures.

[RBD]

Hi Erwin. Worried that this might only be a Dynamica (Maya plugin) problem I tried a quick hack (sorry it's quick and dirty) at BasicDemo.cpp (attached) to test this structure and while I get a structure bounce (and sometimes fall) with a height of 10, it's perfectly solid at a height of 2, unlike in Maya. And if I increase the stepSimulation fixedTimeStep to 1/240 in this BasicDemo, then even with a height of 10 the structure becomes quite solid.

So, even though I get a similar bounce with a tall structure in this BasicDemo it seems that the problem is much worse / exacerbated by the Dynamica Maya Plugin... assuming someone else can confirm the problem in Maya (in case the problem is limited to my machine / compile?). So it does seem to be more a Maya Plugin issue than a general Bullet issue.

Thanks!

[RBD]

Hi Erwin, I've been playing outside of Maya with these Kapla type blocks / planks and it's working very well (still don't know what's wrong with the plugin). The simulation has been running surprisingly solidly even for tall layered structures with a timeStep of 1/24 and a fixedTimeStep 1/240. I'm very impressed with the incredible realistic and robust performance of the physics engine, let alone the speed. Here's a couple of renders I did of my tests:

3000 Kapla type planks simulated with Bullet Physics: YouTube - 3000 Kapla planks in slow motion

It does start to get slow with thousands of rigid bodies. I noticed AMD has recently released their OpenCL SDK. Hopefully I can download it and get started playing with this and the OpenCL branch, perhaps it will allow real-time performance with thousands of rigid bodies?

Thanks.