The dataset viewer is not available for this subset.
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 80, in _split_generators
raise ValueError(
...<2 lines>...
)
ValueError: The TAR archives of the dataset should be in WebDataset format, but the files in the archive don't share the same prefix or the same types.
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 68, 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.
IsaacSkill Dataset
IsaacSkill is a synthetic robot-manipulation dataset containing RGB observations, language task descriptions, object bounding boxes, and dense 2D point tracks. The release contains 2,430 episodes across five skill categories and fifteen task variants.
Dataset summary
| Property | Value |
|---|---|
| Skill categories | 5 |
| Task variants | 15 |
| Episodes per variant | 162 |
| Total episodes | 2,430 |
| Raw RGB frames per episode | 121 |
| Processed RGB/track frames per episode | 60 |
| Tracked points per episode | 1,024 |
| Unpacked size | approximately 185 GiB |
The five integer labels are:
| Label | Skill | Variants |
|---|---|---|
| 0 | Picking and placing | 3 |
| 1 | Pouring | 3 |
| 2 | Pushing | 3 |
| 3 | Sliding-drawer opening | 3 |
| 4 | Hinged-door opening | 3 |
Repository layout
The original dataset contains more than 500,000 small Zarr chunk files. To keep the Hub repository efficient, this release stores each task variant in one uncompressed tar shard. Zarr chunks are already compressed, so applying another compression layer would add CPU cost with little benefit.
data/
Hinge_Opening1.tar
Hinge_Opening2.tar
Hinge_Opening3.tar
Picking&Placing1.tar
Picking&Placing2.tar
Picking&Placing3.tar
Pouring1.tar
Pouring2.tar
Pouring3.tar
Pushing1.tar
Pushing2.tar
Pushing3.tar
Slide_Opening1.tar
Slide_Opening2.tar
Slide_Opening3.tar
metadata/
archive_manifest.jsonl
episodes.jsonl
schema.json
tools/
download_isaacskill_hf.py
Each archive expands to the original task directory:
<task>/
episode_0/
camera_0/rgb # uint8 [121, 480, 640, 3]
rgb_arr # uint8 [60, 256, 256, 3]
bbox_point_tracking_sequence # float32 [1024, 60, 3]
initial_points # float32 [1024, 2]
end_points # float32 [1024, 2]
bbox # [1, 4]
bbox_end # [1, 4]
sample_indices # int32 [60]
label # int64 [1]
task_description # string [1]
...
episode_161/
The first two channels of bbox_point_tracking_sequence are 2D coordinates.
The third channel is binary and is interpreted as the point-validity/visibility
flag. Bounding boxes use the original 640 x 480 image coordinate system. Point
coordinates retain the tracker outputs in the processed image coordinate system.
Download
Download and extract selected task variants with the included helper:
python tools/download_isaacskill_hf.py \
--repo-id huangaoshen/IsaacSkill \
--output-dir ./IsaacSkill_Dataset \
--tasks Pushing1 Pouring1
To use the community mirror for downloads from networks where the official Hub is unavailable:
HF_ENDPOINT=https://hf-mirror.com \
python tools/download_isaacskill_hf.py \
--repo-id huangaoshen/IsaacSkill \
--output-dir ./IsaacSkill_Dataset \
--tasks Pushing1
The public mirror is intended for downloads. Uploads must target the official Hub or a write-capable private gateway.
For a manual single-shard download:
from huggingface_hub import hf_hub_download
archive = hf_hub_download(
repo_id="huangaoshen/IsaacSkill",
repo_type="dataset",
filename="data/Pushing1.tar",
)
print(archive)
After extracting one or more shards, add a root Zarr marker if an application requires an explicit root group:
printf '{"zarr_format": 2}\n' > IsaacSkill_Dataset/.zgroup
Reading an episode with Zarr v2:
import imagecodecs.numcodecs
import zarr
imagecodecs.numcodecs.register_codecs()
episode = zarr.open_group(
"IsaacSkill_Dataset/Pushing1/episode_0",
mode="r",
)
print(episode["task_description"][0])
print(episode["rgb_arr"].shape)
print(episode["bbox_point_tracking_sequence"].shape)
Known-compatible reader dependencies are listed in environment.yml in this
repository.
Dataset index
metadata/episodes.jsonl contains one lightweight record per episode, including
the task variant, integer label, language description, archive name, episode
path, and array shapes. metadata/archive_manifest.jsonl contains archive byte
sizes and SHA-256 checksums.
Intended use
The dataset is intended for research in robot learning, visual representation learning, language-conditioned manipulation, point tracking, and motion prediction. Its synchronized visual, language, bounding-box, and point-track annotations support both end-to-end learning and component-level evaluation.
License
The dataset is released under the license identified in this card's metadata:
cc-by-4.0. The dataset license is separate from the license of any source
code used to process it.
- Downloads last month
- 49