Mass interactions and ways to improve jitter

[SSNTails]

Video #1: The player (dynamic sphere of mass 100) will push objects of a lower mass (convex hull dynamic crate with a mass of 1) through triangle meshes. Aren't static triangle meshes supposed to have infinite mass and therefore the crate should favor clipping through the player sphere instead? Or is there a way to allow the player to stand on top of a crate with a mass lower than the player?

stompdynamic.mp4

Video #2: This shows some barrels (dynamic convex hull) with a mass of 200, while the player is a dynamic sphere of 100 mass. The player can now land and stand on top of the barrels (as desired), but there's some jitter to it.. both the player and the barrel clip inside other objects (between themselves and the triangle mesh) and slowly push themselves into the correct position. Is there a way to reduce that artifacting?

stompdynamic2.mp4

[jrouwe]

Iterative solvers (which is what Jolt uses) are not very good in 'high mass ratios' (where the top object is much heavier than the bottom one). An object that's 100x more heavy than another object will not work (a factor 10x is already a lot). There is no 'priority system' that decides to push 1 object over pushing another object.

W.r.t. the barrel: I don't immediately see what could be going wrong. I'll need some sort of repro for that.

[jrouwe]

Force (Newton) = mass (kg) * acceleration (m/s^2) (and the acceleration = gravity in this case)

[SSNTails]

Adding the appropriate gravity back, I ended up with the same artifacts I originally had. But then I reduced the size of my worlds to 1/5th of their original size (thankfully, I have a global scalar that allows me to do this easily). Suddenly, all of the problems went away.

I'm not sure why this didn't express itself when loading the snapshot into the sample app. Maybe because the default gravity is lower in the sample app? (when I used lower gravity, I didn't get the same effects).

I then tried compiling Jolt using JPH_DOUBLE_PRECISION, with my original world unit size, and it still exhibited my original problematic behavior. So it's not a precision issue - at least one that JPH_DOUBLE_PRECISION can solve.

I don't have any question here.. just some information for you. For now, I'm using the smaller world units and will see if I run into any other troubles. If I try to shrink my world units any smaller, I start getting breakpoints hit in things like when adding a box shape with a width and height of 0.0140000004 . I guess this is too small for Jolt?

[SSNTails]

Aha! I have a repro for you! Check this out -- set the gravity in the sample app to 20, then load this snapshot (the layers are compatible this time). Look at how much things jitter and then easily clip through the static mesh.

fall.zip

SampleAppJitter.mp4

[jrouwe]

It looks like your inertia calculations are off. If in LoadSnapshotTest::Initialize I add:

settings.mOverrideMassProperties = EOverrideMassProperties::CalculateInertia;

to the loop:

for (BodyCreationSettings &settings : scene->GetBodies())

Then it starts behaving normally.

You can visualize the inertia in the samples app by hitting the 'I' key, without the change the inertia looks like that of a tiny box:

with the change it looks like a box that is about the same size as the barrel:

When you use an inertia that is too small, small impulses lead to big angular velocity changes so it continuously bounces back and forth between both sides of the barrel as is visible in your video. As a general rule of thumb: Don't mess with inertia yourself, just let Jolt calculate it for you.

[SSNTails]

That was it!! Glad it was just some small setting I had wrong. I was doing this originally to supply a custom mass value:

bcs.mOverrideMassProperties = JPH::EOverrideMassProperties::MassAndInertiaProvided
bcs.mMassPropertiesOverride.SetMassAndInertiaOfSolidBox(JPH::Vec3(1, 1, 1), mass);

Doing as you said makes everything work:

bcs.mOverrideMassProperties = JPH::EOverrideMassProperties::CalculateInertia;
bcs.mMassPropertiesOverride.SetMassAndInertiaOfSolidBox(JPH::Vec3(1, 1, 1), mass);

[jrouwe]

Note that if you want to save a few cycles, instead of:

bcs.mMassPropertiesOverride.SetMassAndInertiaOfSolidBox(JPH::Vec3(1, 1, 1), mass);

you can do:

bcs.mMassPropertiesOverride.mMass = mass;

as the inertia will no longer be used.