Picking issues with tilesets representing "small" models (and massive logging)

[Julot94]

What happened?

Hello,
Since a recent version (we experienced it with 2.14.1, but are unsure when the change was integrated exactly), we encounter some picking issues and massive warnings in logs with tilesets representing a "small" geometry (a building, typically).

The warning that we get in UE looks like this:

LogPhysics: Warning: Initialising Body : Scale3D is (nearly) zero: Actor: 'xxx' Component: 'xxxx'

And when this happens, it is no longer possible to pick the corresponding primitives (using UE generic trace methods such as UWorld::LineTraceMultiByObjectType).

After comparing with a previous working version, we found out that it was caused by the scaling by CesiumPrimitiveData::positionScaleFactor (currently defined as 1024.0), which scales up the internal coordinates but (logically ;-)) scales down the primitive's transformation, which then becomes "too small" for UE, which then discards the collision mesh (or at least makes the tracing miss some impacts).

Here is a link to a commit dealing with this scaling factor:
22769c2

One possible workaround could consist in making this scaling factor depend on the tileset's dimensions, but I guess @kring would have more context about the issue that made you introduce this factor (it would be bad to get a regression on the cases this factor was fixing...)

Environment

Cesium for Unreal version: 2.14.1
Unreal Engine Version: 5.3
Operating System: Windows 11 Enterprise

Reproduction steps

1. As explained, it depends on the dimensions of the model (I have a few examples coming from iTwin)

...

Supporting evidence

No response

[azrogers]

I've experienced this issue too - not necessarily with picking (I haven't tried), but loading iModel mesh exports seems to produce the above warnings. I don't have a solution, but I can give you some more information on why we're doing that transformation in the first place. Here's the PR where we originally introduced that transformation, along with an in-depth explanation from @kring on why it's needed: #1465

[kring]

@Julot94 (or @azrogers as well, since you mentioned you've seen this), can you help me to reproduce this myself? The positionScaleFactor is meant to fix problems with physics, so it's surprising and concerning that it's creating new ones.

[Julot94]

Hi Kevin, Sure, I will give you an access to the example I have