Ray Leakage in Indoor Scenarios with Sionna RT

[devrimfiratyilmaz]

While working with Sionna RT (v0.17.0) on my indoor scenarios, I have encountered an issue related to ray leakage at specific structural junctions. After converting a 2D floor plan to a 3D PLY format and defining the entire building with the 'mat-itu_marble' material, it was noted that non-white squares were present outside the building on the coverage maps.

I suspected that these anomalies were caused by rays leaking outside, so I placed receivers outside the building to examine the paths in detail. My observations indicated that rays were consistently escaping the building at the junctions where the side walls meet the ceiling/floor.

Despite testing with different conversion methods from 2D to 3D and with the scene_01 of the published work WiSegRT, the issue remained unresolved. This indicated that the rays were still escaping from the building at the same points where the base and the side walls converge.

It seems that there may be an issue with how Sionna handles the intersections between side walls and ceilings. This is causing incorrect indoor/outdoor detection at these points, which is allowing rays to escape.

I would be grateful if you could suggest strategies or modifications to prevent rays from escaping at these junctures, thereby improving the accuracy of simulations.

[merlinND]

Hello @devrimfiratyilmaz,

I believe that OptiX, the library used for ray tracing in Sionna, can guarantee that rays won't leak this way only if the different parts are actually connected (i.e. part of the same mesh). Could you please confirm this is the case?
If not, you might have to very slightly overlap the edges of walls & ceilings to ensure there's no gap due to e.g. limited numerical precision.

[devrimfiratyilmaz]

Thank you for your feedback.

At which version was the mentioned code change implemented? It cannot be the one introduced in v0.18.0, because you can also see the leaks in v0.17.0. We wanted to point this out again because a rollback of the last relevant changes from v0.17.0 to v0.18.0 would apparently not completely fix the problem.

[JackYifan]

@faycalaa May I ask when the new version of Sionna will be released?

[devrimfiratyilmaz]

Hello @faycalaa,

I have just completed testing the scenarios we discussed, now using the newly released version Sionna v0.19. Contrary to the announcement here, while this version addresses the significant ray leakage issue present in v0.18, it does not fully resolve the leakage problem. In fact, the level of leakage observed is roughly similar to what was experienced in v0.17.

As we had expressed earlier, our concerns about rolling back the last relevant changes from v0.17.0 to v0.18.0 not completely addressing the issue seem valid. The reduced, yet persistent, leakage raises questions. As @merlinND mentioned in his response, if the errors are not related to the specific geometry of the scene, what could be causing them?

I would appreciate further insights into whether this might be related to other recent code changes or if additional fixes are planned for future versions. I look forward to your feedback.

[merlinND]

Hello @devrimfiratyilmaz,

Spawning reflected / refracted rays in a way that robustly avoids leakage with all types of geometry (including different scene scales) is far from a trivial problem due to the finite precision of floating point computations. See for example:

• https://www.pbr-book.org/4ed/Shapes/Managing_Rounding_Error#RobustSpawnedRayOrigins
• https://developer.nvidia.com/blog/solving-self-intersection-artifacts-in-directx-raytracing/

The fix mentioned in the v0.19 release notes meant fixing the regression that was introduced in v0.17.
We may introduce even more robust ray spawning algorithms in future versions.

[devrimfiratyilmaz]

Hello @merlinND,

Thank you for the clarification! We look forward to new developments in future versions!