NVIDIA NuRec tutorial is broken

[BaderTim]

Setup

• CARLA version: 0.9.16
• Platform: Ubuntu 24
• Python version: 3.10.20
• GPU: NVIDIA RTX 5000 Ada
• GPU Drivers: 580.126.09

Describe the bug
When running the example_nurec_replay_save_images.py file, the NVIDIA NuRec container won't start due to an unknown calib name.

Steps to reproduce
I strictly followed the tutorial here: https://carla.readthedocs.io/en/0.9.16/nvidia_nurec/

Expected behavior
NuRec container starts up without error and the scene properly loads.

Logs

(carla) user@ZBOOK:~/CARLA_0.9.16/PythonAPI/examples/nvidia/nurec$ python example_nurec_replay_save_images.py --usdz-filename PhysicalAI-Autonomous-Vehicles-NuRec/sample_set/26.02_release/0a18
c5a4-9aca-4efd-b604-c75f3269c502/0a18c5a4-9aca-4efd-b604-c75f3269c502.usdz --move-spectator
/home/user/miniforge3/envs/carla/lib/python3.10/site-packages/pygame/pkgdata.py:25: UserWarning: pkg_resources is deprecated as an API. See https://setuptools.pypa.io/en/latest/pkg_resources.html. The pkg_resources package is slated for removal as early as 2025-11-30. Refrain from using this package or pin to Setuptools<81.
  from pkg_resources import resource_stream, resource_exists
2026-04-12 01:40:48 INFO     CUDA_VISIBLE_DEVICES not set, defaulting to GPU 0
2026-04-12 01:40:48 INFO     Starting container NuRec_clipgt-0a18c5a4-9aca-4efd-b604-c75f3269c502_run_c166eb8e on localhost:46435
2026-04-12 01:40:48 INFO     Waiting for server to start and scene to load...
2026-04-12 01:41:43 ERROR    Critical error detected: [2026-04-11 23:41:43,278][nre.grpc.serve][ERROR] Failed to create backend for clipgt-0a18c5a4-9aca-4efd-b604-c75f3269c502: Unknown calib name='free-pose-calib'.

I also tried to modify the script to start the server without --test-scenes-are-valid:

(carla) user@ZBOOK:~/CARLA_0.9.16/PythonAPI/examples/nvidia/nurec$ python example_nurec_replay_save_images.py --usdz-filename PhysicalAI-Autonomous-Vehicles-NuRec/sample_set/26.02_release/0a18
c5a4-9aca-4efd-b604-c75f3269c502/0a18c5a4-9aca-4efd-b604-c75f3269c502.usdz --move-spectator
/home/user/miniforge3/envs/carla/lib/python3.10/site-packages/pygame/pkgdata.py:25: UserWarning: pkg_resources is deprecated as an API. See https://setuptools.pypa.io/en/latest/pkg_resources.html. The pkg_resources package is slated for removal as early as 2025-11-30. Refrain from using this package or pin to Setuptools<81.
  from pkg_resources import resource_stream, resource_exists
2026-04-12 09:50:43 WARNING  Existing container NuRec_clipgt-0a18c5a4-9aca-4efd-b604-c75f3269c502_run_69c8e993 verification failed
2026-04-12 09:50:43 INFO     Killing old NuRec containers with same image: ['NuRec_clipgt-0a18c5a4-9aca-4efd-b604-c75f3269c502_run_69c8e993']
2026-04-12 09:50:45 INFO     CUDA_VISIBLE_DEVICES not set, defaulting to GPU 0
2026-04-12 09:50:45 INFO     Starting container NuRec_clipgt-0a18c5a4-9aca-4efd-b604-c75f3269c502_run_fc02f48e on localhost:46435
2026-04-12 09:50:45 INFO     Waiting for server to start and scene to load...
2026-04-12 09:51:32 ERROR    Critical error detected: [2026-04-12 07:51:32,621][nre.grpc.serve][ERROR] Failed to create backend for clipgt-0a18c5a4-9aca-4efd-b604-c75f3269c502: Unknown calib name='free-pose-calib'.
2026-04-12 09:51:32 ERROR    Timeout waiting for server to be ready
2026-04-12 09:51:33 WARNING  Scene validation failed due to unsupported calibration metadata; retrying startup without --test-scenes-are-valid.
2026-04-12 09:51:33 INFO     Starting container NuRec_clipgt-0a18c5a4-9aca-4efd-b604-c75f3269c502_run_0716e4ea on localhost:46435
2026-04-12 09:51:33 INFO     Waiting for server to start and scene to load...
2026-04-12 09:52:15 INFO     Server is ready! gRPC service is serving without scene validation.
Traceback (most recent call last):
  File "/home/user/CARLA_0.9.16/PythonAPI/examples/nvidia/nurec/example_nurec_replay_save_images.py", line 333, in <module>
    main()
  File "/home/user/CARLA_0.9.16/PythonAPI/examples/nvidia/nurec/example_nurec_replay_save_images.py", line 286, in main
    with NurecScenario(
  File "/home/user/CARLA_0.9.16/PythonAPI/examples/nvidia/nurec/nurec_integration.py", line 720, in __enter__
    self.renderer = NurecRenderer(
  File "/home/user/CARLA_0.9.16/PythonAPI/examples/nvidia/nurec/nurec_integration.py", line 359, in __init__
    self._init_grpc()
  File "/home/user/CARLA_0.9.16/PythonAPI/examples/nvidia/nurec/nurec_integration.py", line 382, in _init_grpc
    self.client_service.get_available_cameras(
  File "/home/user/miniforge3/envs/carla/lib/python3.10/site-packages/grpc/_channel.py", line 1181, in __call__
    return _end_unary_response_blocking(state, call, False, None)
  File "/home/user/miniforge3/envs/carla/lib/python3.10/site-packages/grpc/_channel.py", line 1006, in _end_unary_response_blocking
    raise _InactiveRpcError(state)  # pytype: disable=not-instantiable
grpc._channel._InactiveRpcError: <_InactiveRpcError of RPC that terminated with:
        status = StatusCode.UNKNOWN
        details = "Exception calling application: Unknown calib name='free-pose-calib'."
        debug_error_string = "UNKNOWN:Error received from peer  {grpc_message:"Exception calling application: Unknown calib name=\'free-pose-calib\'.", grpc_status:2, created_time:"2026-04-12T09:52:23.333013954+02:00"}"

[kmsngh]

You need to use the 25.07 release of the Physical AIAV NuRec dataset.

[bhanu-iav]

are they planning to support NuReC and Nvidia in UE5?

[Chinese-Coding]

Did you solve this problem? I encountered the same error even though I followed the tutorial.

By the way, I found that the dataset size is more than 200 GB. It may be larger than 253 GB, because I stopped downloading it when I saw that the size was already over 200 GB. However, the tutorial says it is about 200 GB. Is this because NVIDIA updated the dataset?

[Chinese-Coding]

Did you solve this problem? I encountered the same error even though I followed the tutorial.

By the way, I found that the dataset size is more than 200 GB. It may be larger than 253 GB, because I stopped downloading it when I saw that the size was already over 200 GB. However, the tutorial says it is about 200 GB. Is this because NVIDIA updated the dataset?

I tried the 25.07 release. It worked.

[buhexb]

Did you solve this problem? I encountered the same error even though I followed the tutorial.
By the way, I found that the dataset size is more than 200 GB. It may be larger than 253 GB, because I stopped downloading it when I saw that the size was already over 200 GB. However, the tutorial says it is about 200 GB. Is this because NVIDIA updated the dataset?

I tried the 25.07 release. It worked.

Are you currently troubleshooting issues with version 26.02? I can only use solutions compatible with version 25.07.

[im-hashim]

Did you solve this problem? I encountered the same error even though I followed the tutorial.

By the way, I found that the dataset size is more than 200 GB. It may be larger than 253 GB, because I stopped downloading it when I saw that the size was already over 200 GB. However, the tutorial says it is about 200 GB. Is this because NVIDIA updated the dataset?

25.07 full download for me was about 1.5TB using the included nurec setup script.

However all it needs is about 2 GB of USDZ files which can be searched using keywords on NGC and downloaded individually using cosmos cookbooks or Hugging face API directly.

[Chinese-Coding]

Are you currently troubleshooting issues with version 26.02? I can only use solutions compatible with version 25.07.

No. I am just using 25.07 and do not want to figure it.

[Chinese-Coding]

25.07 full download for me was about 1.5TB using the included nurec setup script.

However all it needs is about 2 GB of USDZ files which can be searched using keywords on NGC and downloaded individually using cosmos cookbooks or Hugging face API directly.

That is a good news. Maybe the document of carla should update.

[buhexb]

Are you currently troubleshooting issues with version 26.02? I can only use solutions compatible with version 25.07.

No. I am just using 35.07 and do not want to figure it.

Thank you for your reply. Are you using a self-built dataset or the official dataset? If you reconstruct USDZ files with your own data, it seems the official NuRec reconstruction image for version 25.07 is not available.

[im-hashim]

25.07 full download for me was about 1.5TB using the included nurec setup script.
However all it needs is about 2 GB of USDZ files which can be searched using keywords on NGC and downloaded individually using cosmos cookbooks or Hugging face API directly.

That is a good news. Maybe the document of carla should update.

Yes I agree. and it is scattered all over place. But sure is much better now than when released mid last year. as many users already put NuRec to test.

Here's a good place to search:
https://build.nvidia.com/nvidia/cosmos-dataset-search

and HF already lists how to download based on labels (section: Downloading usdz based upon categories in labels.json)
https://huggingface.co/datasets/nvidia/PhysicalAI-Autonomous-Vehicles-NuRec#downloading-data

What I personally use is a python notebook similar to:
https://github.com/NVlabs/alpamayo/blob/main/notebooks/inference.ipynb

Then, if you want to reconstruct USDZ files with your own data, NCORE would be a light weight addition to your workflow
https://github.com/NVIDIA/ncore

I hope this helps!

[Chinese-Coding]

Thank you for your reply. Are you using a self-built dataset or the official dataset? If you reconstruct USDZ files with your own data, it seems the official NuRec reconstruction image for version 25.07 is not available.

Yes, I intend to conduct the reconstruction using my own dataset. Currently, I am utilizing NVIDIA's official nurec dataset solely for the purpose of feasibility validation.

I think I would like to discuss your mentioned problem after I test reconstruction with my own dataset.

[im-hashim]

Are you currently troubleshooting issues with version 26.02? I can only use solutions compatible with version 25.07.

No. I am just using 35.07 and do not want to figure it.

Thank you for your reply. Are you using a self-built dataset or the official dataset? If you reconstruct USDZ files with your own data, it seems the official NuRec reconstruction image for version 25.07 is not available.

Tracing back the download script. 25.07 used the following:
nvidia-nurec-grpc/0.2.0

latest is:
nvcr.io/nvidia/nre/nre-ga:26.02.00
If your self built data set was using NCORE v3 and v4, they should be compatible with this container.