Datasets:

jandan138
/

GRScenes-portable

Exception:    SplitsNotFoundError
Message:      The split names could not be parsed from the dataset config.
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 286, in get_dataset_config_info
                  for split_generator in builder._split_generators(
                                         ~~~~~~~~~~~~~~~~~~~~~~~~~^
                      StreamingDownloadManager(base_path=builder.base_path, download_config=download_config)
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  )
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/webdataset/webdataset.py", line 81, in _split_generators
                  first_examples = list(islice(pipeline, self.NUM_EXAMPLES_FOR_FEATURES_INFERENCE))
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/webdataset/webdataset.py", line 32, in _get_pipeline_from_tar
                  fs: fsspec.AbstractFileSystem = fsspec.filesystem("memory")
                                                  ~~~~~~~~~~~~~~~~~^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/fsspec/registry.py", line 302, in filesystem
                  cls = get_filesystem_class(protocol)
                File "/usr/local/lib/python3.14/site-packages/fsspec/registry.py", line 239, in get_filesystem_class
                  raise ValueError(f"Protocol not known: {protocol}")
              ValueError: Protocol not known: memory
              
              The above exception was the direct cause of the following exception:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/config/split_names.py", line 71, in compute_split_names_from_streaming_response
                  for split in get_dataset_split_names(
                               ~~~~~~~~~~~~~~~~~~~~~~~^
                      path=dataset,
                      ^^^^^^^^^^^^^
                      config_name=config,
                      ^^^^^^^^^^^^^^^^^^^
                      token=hf_token,
                      ^^^^^^^^^^^^^^^
                  )
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 340, in get_dataset_split_names
                  info = get_dataset_config_info(
                      path,
                  ...<6 lines>...
                      **config_kwargs,
                  )
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 291, in get_dataset_config_info
                  raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
              datasets.inspect.SplitsNotFoundError: The split names could not be parsed from the dataset config.

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

GRScenes Portable v1.0.0-rc1

Release candidate. Complete Windows GUI visual acceptance and actual Linux Isaac Sim validation are still pending. Reproducible downloads must use the immutable v1.0.0-rc1 tag, not the mutable main branch. A refs/pr/* view is public upload staging and must not be treated as a release.

GRScenes Portable is a GRScenes laboratory-authorized portable release maintained by a project member. It preserves the original scene and asset organization while removing private /cpfs material paths, restoring two transport-damaged USD layers from the laboratory canonical copy, and closing a small set of missing material dependencies. It is a compatibility repair, not a new dataset.

The payload contains 99 environments (69 home and 30 commercial), their deduplicated asset library, and the required MDL/material resources. The release contains 394,508 payload files and 291,181,357,750 bytes before archive compression. All authored USD asset paths are intended to resolve after extraction at an arbitrary Windows or Linux location. Isaac Sim, Omniverse Kit, and NVIDIA drivers are not included.

For a Chinese quick start, see README_zh-CN.md.

License and access terms

The released GRScenes data is licensed under CC BY-NC-SA 4.0. Attribution, non-commercial use, and ShareAlike requirements apply. Preserve LICENSE-GRSCENES-CC-BY-NC-SA-4.0.txt and ATTRIBUTION.md when sharing adaptations.

The upstream project also asks users to complete the GRScenes-100 Dataset Access User Agreement. This public mirror does not waive that requirement. See the official GRScenes Dataset Card and InternUtopia assets documentation.

Requirements

Approximately 291 GB (271 GiB) for the extracted payload, plus compressed shards, Hugging Face cache, and a safety margin during installation.
Python 3.10 or newer for downloading and verification.
huggingface_hub with hf_xet, plus zstandard; tested versions are pinned in requirements-download.txt.
NVIDIA Isaac Sim 6.x to render and run USD stages with MDL materials.

Download, verify, and extract

Install the downloader requirements from this repository, then run the public interface pinned to the release-candidate tag:

python -m pip install -r requirements-download.txt
python tools/download_and_extract.py \
  --repo-id jandan138/GRScenes-portable \
  --revision v1.0.0-rc1 \
  --output /data/grscenes

The downloader retrieves all 21 tar.zst shards through Hugging Face/Xet, verifies every shard SHA-256, rejects unsafe archive members, extracts with per-file atomic replacement, and automatically runs both payload checks. It refuses to overwrite a non-empty output directory.

--cache-dir PATH    place the Hugging Face cache on a selected volume
--download-only     verify downloads without extracting
--extract-only      extract an already downloaded local shard set
--resume            reuse only files whose path, size, and SHA-256 match
--keep-shards       retain materialized shard copies after success

Manual verification after extraction:

python /data/grscenes/tools/verify_dataset.py --dataset /data/grscenes --fast
python /data/grscenes/tools/verify_dataset.py --dataset /data/grscenes --keyfiles

Isaac Sim usage

On Linux, add the bundled MDL root only for the current shell:

source /data/grscenes/tools/setup_isaacsim_env.sh

Then open a scene such as:

/data/grscenes/GRScenes100/home/MV7J6NIKTKJZ2AABAAAAADA8_usd/layout.usd

Validate every layout using the Python environment shipped with Isaac Sim:

/path/to/isaac-sim/python.sh \
  /data/grscenes/tools/validate_stages.py \
  --dataset /data/grscenes \
  --output-json /data/grscenes/stage-validation.json \
  --output-csv /data/grscenes/stage-validation.csv

On Windows, use Isaac Sim's python.bat with Windows paths. A generic PyPI usd-core environment is not a substitute for MDL/Kit validation.

Payload and archive layout

dataset/
├── GRScenes100/       # 99 layout.usd entry points
├── GRScenes_assets/   # deduplicated object and layout assets
├── Material/          # MDL modules and texture resources
├── tools/             # verification and Isaac Sim helper scripts
├── MANIFEST.payload.csv
├── MANIFEST.keyfiles.sha256
├── PORTABLE_RELEASE.json
├── README.md
└── ATTRIBUTION.md

The Hub payload is stored in 21 deterministic archives: one scenes shard, 18 hash-stable asset shards, and two hash-stable Material shards. Members are rooted at dataset/ and contain only regular files and directories. The release does not rely on symlinks, junctions, hardlinks, or device files.

Portable repair and material disclosure

The release replaces the approved private Material prefix with layer-relative POSIX-style AssetPaths and localizes one authored Twinbru URL. Layout layers were not rewritten by the portability pass; the two transport-damaged layouts were restored byte-for-byte from the laboratory canonical source before release.

Seven formerly missing MDL modules and their required runtime closure came from official NVIDIA distribution locations and passed Isaac Sim 6.0.1 MDL validation. Two compatibility fallbacks are disclosed:

Material/mdl/Graphical_White_Jacquard_FL_3075963.mdl is a neutral, texture-free 50% gray MDL fallback. The official Twinbru candidate did not reliably terminate registration under the full dataset MDL search root. The fallback preserves the USD-required symbol and has SHA-256 9af2cddf57e5458c29f31601a6a386f4264a9b3f782f7fd33dfd07a98684e834.
Material/mdl/textures/eyes03A.jpg is a true-JPEG 64×64 uniform 50% gray compatibility texture. It does not reconstruct the missing eye texture and has SHA-256 94b2f9daa78a89a37eced1a78698e54f9201ad0908106f52d03d760fbc1d552f.

No MDL-to-UsdPreviewSurface conversion was performed. Software without NVIDIA MDL support may display geometry without complete materials.

Validation status and known GUI limitations

The immutable tag records machine-readable details in PORTABLE_RELEASE.json. The non-GUI release-candidate gates are:

Windows Isaac Sim 6.0.1 structural validation: 99/99 layouts passed; unresolved references, outside-dataset dependencies, and Stage hard errors were zero.
MDL validation: 1,574/1,574 unique bindings passed with zero failed batches.
Local and Hugging Face PR archive round-trips must pass shard hashes, payload manifests, Windows Stage/MDL validation, and WSL structure validation before the PR is merged and tagged.
WSL Ubuntu 22.04 checks exact filename case, relative-path containment, absence of link/reparse dependencies, and both manifest modes.

Complete GUI visual acceptance is not claimed for v1.0.0-rc1:

the strict targeted GUI survey did not reach PASS because of unapproved Kit/RTX/SdfPath/normal-primvar warnings and memory limits;
Scene 10 produced a black automated frame;
Scene 12 later produced a dark frame with warnings;
Scene 13 remained memory-blocked and has no accepted screenshot;
the neutral eyes03A.jpg fallback has not been confirmed from a front-facing character view.

These observations do not mean the structural or MDL gates failed. Complete Windows GUI acceptance and at least one real Linux Isaac Sim 6.x recipient validation are required before a stable v1.0.0 release.

Upstream and citation

@inproceedings{grutopia,
  title={GRUtopia: Dream General Robots in a City at Scale},
  author={Wang, Hanqing and Chen, Jiahe and Huang, Wensi and Ben, Qingwei and Wang, Tai and Mi, Boyu and Huang, Tao and Zhao, Siheng and Chen, Yilun and Yang, Sizhe and Cao, Peizhou and Yu, Wenye and Ye, Zichao and Li, Jialun and Long, Junfeng and Wang, ZiRui and Wang, Huiling and Zhao, Ying and Tu, Zhongying and Qiao, Yu and Lin, Dahua and Pang, Jiangmiao},
  year={2024},
  booktitle={arXiv}
}

Versioning and support

v1.0.0-rc1 is immutable. A payload correction will use a new tag such as v1.0.0-rc2; this tag and its manifests will never be moved or overwritten. Stable v1.0.0 will be published through a separate reviewed tree after the remaining GUI and Linux gates pass. Report problems with the scene path, platform, Isaac Sim version, validator JSON, and failing manifest path without attaching licensed payload files to public issues.

Downloads last month: 55

Total file size:

170 GB

Paper for jandan138/GRScenes-portable

GRUtopia: Dream General Robots in a City at Scale

Paper • 2407.10943 • Published Jul 15, 2024 • 25