Rigidbody linear damping has non-intuitive behavior around 1.0

[Bauxitedev]

Godot version:
3.0.2 Mono

OS/device including version:
KDE Neon

Issue description:

When adjusting the linear damping of a rigid body in the range 0.98 to 1.02, the difference between 0.98 and 0.99 is marginal, but going from 0.99 to 1 suddenly causes a huge spike in the amount of dampening applied. Then going from 1 to 1.01 has a marginal effect again.

This is really annoying because for one of my projects, a damping on 0.99 was not enough, but a damping of 1.0 is way too much. There's no way to get a value inbetween. (0.999 is rounded to 1 in the editor)

This is using the DEFAULT physics engine, by the way.

Demonstration:

Might be related: #14650

Steps to reproduce:

1. Create a Rigidbody, set linear damping to 0.99 and run the project. Observe behaviour.
2. Set its linear damping to 1.0 and observe it suddenly falls much more slowly.

Minimal reproduction project:
dampingbug.zip

[mhilbrunner]

cc @AndreaCatania

[AndreaCatania]

The damping seems applied correctly. I don't know if this is a feature or a bug. I need to open an issue to the Bullet Physics github to assert it

[jitspoe]

From looking at the code, it seems damping is designed to be in the [0-1) range. Here's what it does in bullet:

m_linearVelocity *= btPow(btScalar(1) - m_linearDamping, timeStep);

So if your damping is 1, it should actually have no movement, and > 1 should... I guess move it backward? I guess the bigger problem is that you can't enter values with more than 2 decimal places, since it seems like you might need to have a really-close-to-1 value. I guess you could add a script to your object and set the value from that so you have more digits available?

[akien-mga]

Maybe damping should be exposed with a different scale? You need a ridiculous precision to be able to use a damping value as defined by the bullet equation:

(1-0,99)^0.016667
= 0,92611730648509949065

(1-0,999)^0.016667
= 0,89124888595498392525

(1-0,99999)^0.016667
= 0,82540101766847533075

(1-0,999999)^0.016667
= 0,79432457671599994306

(1-0,9999999)^0.016667
= 0,7644181672531883877

(1-0,999999999999)^0.016667
= 0,63095153317505247888

(1-0,9999999999999999)^0.016667
= 0,54116288077988569416

(1-0,9999999999999999999)^0.016667
= 0,48231081461526290744

(1-0,999999999999999999999999)^0.016667
= 0,39809983721594934885

[AndreaCatania]

Hello @erwincoumans, in the PR #37314 we are considering to change the damping behaviour in order to be more controllable. See: #19182 (comment)

However, By reading this https://code.google.com/archive/p/bullet/issues/74 is possible to see that the formula was changed and it seems the cause of this weird behaviour.

Is this behaviour expected?

[erwincoumans]

Damping us expected to be in the range [0..1], best closer to 0 not 1. I would implement custom damping in a pretick callback. Or just implement your own version behind an #ifdef, happy to add merge it in the Bullet repo (but leave default as-is, I don't want to break the API).

[AndreaCatania]

@erwincoumans Thanks for the replay, Do you think it's expected this behavioural differences between a body with damping 0.99 and 1.0?

We would expect something more smooth so would be useful understand if there is a problem or why it's expected to work as so.