Modified sample_pointcloudprocessing: fused point cloud OK, but range image always black / depth = 0

Hi,

I’m using DRIVE OS 6.0.10 / DriveWorks 5.20.24 on DRIVE Orin P3710.

I modified sample_pointcloudprocessing to work with my setup:

  • 4x Hesai AT128 via custom lidar.custom plugin
  • custom rig
  • point cloud stitching/fusion similar to the original sample

The important part:

  • The original, unmodified sample works on my system.
  • My modified version renders the fused point cloud correctly.
  • Only the range image stays completely black.

What works:

  • all LiDARs decode correctly
  • fused/stiched point cloud looks plausible
  • rendering pipeline is fine (I verified this with a simple gradient test)

What does not work:

  • dwPCRangeImageCreator_process() returns no error
  • but m_stitchedDepthMap3D, m_stitchedDepthImage, and the host copy contain only zeros
  • all output values are finite, but depth/range is always 0

Debug output example:

  • stitched input cloud: valid ranges, plausible XYZ values
  • depth map output: all (0,0,0), range = 0
  • range image: all pixels = 0

What I already tried:

  • recorded LiDAR data instead of live data
  • no CAN / no DBC
  • very permissive clipping settings (huge OBB, elevation ±90°)

Result is always the same.

So it seems the issue happens inside the range image creation step, before visualization.

My main questions:

  1. What conditions can cause dwPointCloudRangeImageCreator to output only zeros although the input cloud is valid?
  2. Are there specific assumptions about the input point cloud format/order/layout that may be violated after modifying the sample?
  3. Does the range image creator expect a specific sensor model or organized cloud structure that may not hold for Hesai/custom fused clouds?

If helpful, I can also share:

  • a minimal rig
  • recorded LiDAR data
  • the relevant modified code section

Any hints on what exactly to verify next would be very appreciated.
Thanks!

Dear @Jis ,
Let me check with engineering team on your queries and get back to you if need any additional details.

  1. This behaviour is observed when using spinning lidars too. We get range image pixels as 0.

Dear @Jis ,
I notice range image creation is not working with point cloud sample in DW 5.20. It is a known issue and fix not available in DW 5.20

@SivaRamaKrishnaNV

Thanks for the clarification.

The strange thing is that I am sure I had the range image working before — I definitely remember seeing it on my setup.

I have now checked all of my old Docker containers, but I cannot reproduce that state anymore. So I assume it was not the final prebuilt sample from /usr/local/driveworks/bin , but one of the binaries I had previously built from the NVIDIA source code.

Those old mounted build folders under /home/nvidia/build-x86_64-linux-gnu are unfortunately already deleted, so I cannot verify the exact build anymore.

This makes me think that the issue is not fully deterministic / not 100% reproducible, or at least depends on a specific build state.

Do you know whether this DW 5.20 issue should always occur, or only in certain configurations?

Thanks!

I notice it with default point cloud sample shipped in DW 5.20.