apply_torque_impulse() problem with Bullet

[HugoSlt]

Godot version:
Godot 3.1 #2928091

OS:
Ubuntu 18.04

Issue description:
With a simple pendulum (a RigidBody attached with a HingeJoint), if we apply
a single torque impulse, we expect the pendulum to start oscillating.
It correctly does so with GodotPhysics but it doesn't with Bullet. With
Bullet the pendulum starts rotating at constant speed.

Here is a video of the problem, first Bullet, then GodotPhysics.

Steps to reproduce:
Create a RigidBody attached to a HingeJoint, apply a single torque impulse on it with apply_torque_impulse(). Switch between Bullet and GodotPhysics in the project settings
to observe the different behaviors.

Minimal reproduction project:
apply_torque.zip

[akien-mga]

CC @AndreaCatania

[TGRCdev]

Currently testing on a build of master branch, commit 2db4942 on Ubuntu 18.04

The reproduction project gives the same result as the above gif. However, when I tried to recreate the project, I didn't encounter the bug:

Here's my project recreation: BulletJoints.zip

[HugoSlt]

I confirm TGRCdev observation, same here : apply_torque.zip has the problem and BulletJoints.zip doesn't.

I think I found where the issue comes from : the centers of our RigidBodies are not at the same place.

In BulletJoints.zip, if you move the children of the Box RigidBody (i.e MeshInstance & CollisionShape) so that when you select the Box, the center of the Box is aligned with the pivot point like this :

then you get the same problem as in apply_torque.zip.

[TGRCdev]

Dang, nice find. Looks like Bullet uses the RigidBody node's position, while GodotPhysics uses the CollisionShape node's position.

[HugoSlt]

So actually it's more a problem of convention than an implementation issue. Maybe those details should be clearly documented somewhere in the docs, especially those points :

• They are multiple issues about the center of mass : Center of mass offset for RigidBody and RigidBody2D. #7136, RigidBody with polygon shape has weird physics. #8638, Center of mass for RigidBody2D #12353 and also this commit Fixed inertia tensor computation and center of mass #7426.
  So it would be nice to sum up the current situation about the center of mass as in Godot 3.1 (i.e can Godot compute the center of mass for us and how, or will it just take the origin of the mesh ? Is it the same for both RigidBody2D and RigidBody (3D) ? Is it the same for both Bullet and GodotPhysics ? etc...)

• Precise whereas Bullet/GodotPhysics use the RigidBody/CollisionShape origin as center of mass

[TGRCdev]

So I've looked a bit further into the differences in Bullet and Godot's center_of_masses, and I have a modified apply_impulse.zip to show some of my findings.

apply_torque_modified.zip

Green Ball = RigidBody's origin
Red Ball = Center of mass
Blue Square = StaticBody

Left Slider = CollisionShape y translation
Top Slider = StaticBody y translation
Right Slider = RigidBody y translation

value of center_of_mass pulled from PhysicsDirectBodyState

---

GodotPhysics

• GDPhysics' center of mass is described in positional offset from the RigidBody node. (Took me a while to figure this out.)
• The center of mass is located at the center of all CollisionShapes under the RigidBody node.
• Multiple CollisionObjects modify the center of mass
• Changes to RigidBody.translation are completely ignored

---

Bullet

(The red ball is hidden under the green ball, because the center of mass is the RigidBody node's origin)

• Bullet's center of mass is described in global coordinates
• The center of mass is located at the RigidBody's origin
• Multiple CollisionObjects do not modify the center of mass, it always stays at RigidBody node's origin
• Changes to RigidBody.translation briefly moves the RigidBody node away from the HingeJoint, but it snaps back after a few frames.

---

Overall, I think we should follow Bullet's center of mass policy. Calculated center of mass might not always be accurate (i.e. an axe with a metal head and a wooden handle), and moving CollisionShapes and MeshInstances is much simpler.

[AndreaCatania]

Yes I can confirm that in Bullet is possible to "Customize", by moving shapes, the center of mass. Would be awesome if someone could write it in the documentation.

[HugoSlt]

Yes it's nice to be able to set the center of mass where we want by adjusting the relative position of the RigidBody/CollisionShapes.
Should we open a new issue on godot-docs to remember documenting this and close this issue ?

Edit : I've opened this issue as a remainder to talk about it in the docs