The dataset viewer is not available for this split.
Error code: StreamingRowsError
Exception: ValueError
Message: Expected object or value
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 290, in _generate_tables
pa_table = paj.read_json(
io.BytesIO(batch), read_options=paj.ReadOptions(block_size=block_size)
)
File "pyarrow/_json.pyx", line 364, in pyarrow._json.read_json
File "pyarrow/error.pxi", line 155, in pyarrow.lib.pyarrow_internal_check_status
File "pyarrow/error.pxi", line 92, in pyarrow.lib.check_status
raise convert_status(status)
pyarrow.lib.ArrowInvalid: JSON parse error: Column() changed from object to string in row 0
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "/src/services/worker/src/worker/utils.py", line 147, in get_rows_or_raise
return get_rows(
dataset=dataset,
...<4 lines>...
column_names=column_names,
)
File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
return func(*args, **kwargs)
File "/src/services/worker/src/worker/utils.py", line 127, in get_rows
rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
File "/src/services/worker/src/worker/utils.py", line 483, in safe_iter
yield from ds.decode(False) if ds.features else ds
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2951, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2461, in __iter__
for key, pa_table in self._iter_arrow():
~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2486, in _iter_arrow
for key, pa_table in self.ex_iterable._iter_arrow():
~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 547, in _iter_arrow
for key, pa_table in iterator:
^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 430, in _iter_arrow
for key, pa_table in self.generate_tables_fn(**gen_kwags):
~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 304, in _generate_tables
batch = json_encode_fields_in_json_lines(original_batch, json_field_paths)
File "/usr/local/lib/python3.14/site-packages/datasets/utils/json.py", line 111, in json_encode_fields_in_json_lines
examples = [ujson_loads(line) for line in original_batch.splitlines()]
~~~~~~~~~~~^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/utils/json.py", line 20, in ujson_loads
return pd.io.json.ujson_loads(*args, **kwargs)
~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
ValueError: Expected object or valueNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
OpenArm Human RLHND PRQ / FP48 — Huiwon WAM-v1 subset
The default release is now exactly the 1,187 episodes in
/data/huiwon/data/openarm_wam_v1/human_as_openarm28/.
It contains 321,095 reference frames. The source selection, episode IDs, task
indices, and prompt strings match that dataset's metadata. Do not infer the
selection from the older openarm_dataset_192x256 YAML sampling ratios.
| Dataset route | Reference subset | Episodes | Frames | FPS |
|---|---|---|---|---|
rlwrld_human_lerobot |
rlwrld_human_lerobot |
472 | 68,065 | 20 |
open_air_fryer |
openarm_validation_v2_junhyeong/open_air_fryer |
52 | 10,751 | 20 |
close_air_fryer |
openarm_validation_v2_junhyeong/close_air_fryer |
52 | 12,518 | 20 |
left_hand_box_white_container |
openarm_validation_v2_junhyeong/left_hand_box_white_container |
101 | 25,362 | 20 |
anyh2r_prq30 |
anyh2r |
510 | 204,399 | 30 |
| Total | 1,187 | 321,095 | mixed |
The seven skill categories are PnP, PnP x 2, Open, Close, Pour, Move, and Flip.
There are 18 distinct instruction strings; skill categories are not instruction
IDs. The AnyH2R stir subset is excluded. Human PnP IDs
31, 67, 93, 256, 313, 409, and 458 are excluded, exactly as in the reference.
Episode IDs are deliberately not compacted. The box directory's historical
left_hand name does not change its right-hand instruction.
Prompts and frame correspondence
meta/tasks.jsonl, meta/episodes.jsonl, and every parquet task_index agree
with the reference, including capitalization, punctuation, and wording. The
older Left hand: None. Right hand: prefix is absent where absent in WAM-v1.
Neither synonym substitution nor an additional language template is applied.
AnyH2R is retargeted from existing full-rate RLHND observations at native 30 Hz.
It is not an interpolation of the earlier 20 Hz cache. Four other subsets
retain their 20 Hz reference row timelines and already validated geometric
labels. Each frame_maps/*.jsonl identifies exact reference RGB frames; the
original high-resolution frame map is a separate field and must not be applied
again to the already sampled reference video. FP48 always uses the right-camera
coordinate system. AnyH2R's video bindings retain both left and right views;
the other subsets retain the right view.
What is included
Offline PRQ labels, FP48 targets, validity masks, freshly recomputed per-subset state/action statistics, frame maps, prompt tables, reference metadata hashes, and numerical validation. No RGB videos, RLHND weights, MANO meshes, contact outputs, or raw human data are uploaded. No new RLHND inference was performed for this subset release.
PRQ/FP48 geometry retains the rlhnd-rightcam-inspire-palm-v1 convention and
existing robot-derived FP48 normalization. The four right-hand subsets use
15D absolute-camera PRQ with next-frame actions; AnyH2R uses bimanual 30D PRQ
with a shared first-valid-right-wrist reference and same-frame actions.
Read meta/label_contract.json rather than assuming these actions have the
same coordinate or temporal convention. Invalid coordinates remain masked.
This is a geometry-label companion for the paper's data selection, not the original paper's 28D action tensor. The reference FRED dataset has dummy zero-filled 28D human state/actions and masks human action loss. Using these PRQ labels for action or alignment supervision changes that training recipe. The old pretrained HaWoR-target models are not claimed to be label-compatible. Numerical checks do not certify visual accuracy, depth, or rollout success.
Download and bind existing reference videos
hf download junbrro/openarm-human-rlhnd-prq-fp48 \
--repo-type dataset --local-dir /your/cache/openarm-human-rlhnd
python /your/cache/openarm-human-rlhnd/prepare_labels.py \
--output /your/new/dataset/rlhnd-wam-v1 \
--video-root /data/huiwon/data/openarm_wam_v1/human_as_openarm28
Use a new output directory. For numerical labels without local videos, pass
--labels-only instead of --video-root. The helper verifies the selected
archive and every member hash before binding paths. label_routes.json gives
the dataset/cache routes. This root contains no extra box/ball/towel view.
latest.json is the authoritative default-release pointer. Old immutable
archives may remain in releases/ for reproducibility; they are not part of
this selection. In particular, release af6b5b6aefcc22db was the previous
1,614-episode expanded MT corpus and must not be mixed into this paper subset.
Pin the HF commit or release ID when sharing an experiment.
- Downloads last month
- 72