The dataset viewer is not available for this split.
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
norm_stats: struct<action.arm.position: struct<min: list<item: double>, max: list<item: double>, mean: list<item (... 1062 chars omitted)
child 0, action.arm.position: struct<min: list<item: double>, max: list<item: double>, mean: list<item: double>, std: list<item: d (... 158 chars omitted)
child 0, min: list<item: double>
child 0, item: double
child 1, max: list<item: double>
child 0, item: double
child 2, mean: list<item: double>
child 0, item: double
child 3, std: list<item: double>
child 0, item: double
child 4, count: list<item: int64>
child 0, item: int64
child 5, q01: list<item: double>
child 0, item: double
child 6, q10: list<item: double>
child 0, item: double
child 7, q50: list<item: double>
child 0, item: double
child 8, q90: list<item: double>
child 0, item: double
child 9, q99: list<item: double>
child 0, item: double
child 1, action.effector.position: struct<min: list<item: double>, max: list<item: double>, mean: list<item: double>, std: list<item: d (... 158 chars omitted)
child 0, min: list<item: double>
child 0, item: double
child 1, max: list<item: double>
child 0, item: double
child 2, mean: list<item: double>
child 0, item: double
child 3, std: list<item: double>
child 0, item: double
child 4, count: list<item: int64>
...
d 3, observation.state.effector.position: struct<min: list<item: double>, max: list<item: double>, mean: list<item: double>, std: list<item: d (... 158 chars omitted)
child 0, min: list<item: double>
child 0, item: double
child 1, max: list<item: double>
child 0, item: double
child 2, mean: list<item: double>
child 0, item: double
child 3, std: list<item: double>
child 0, item: double
child 4, count: list<item: int64>
child 0, item: int64
child 5, q01: list<item: double>
child 0, item: double
child 6, q10: list<item: double>
child 0, item: double
child 7, q50: list<item: double>
child 0, item: double
child 8, q90: list<item: double>
child 0, item: double
child 9, q99: list<item: double>
child 0, item: double
episodes: struct<clean: int64, A_color_geometry: int64, B_noise: int64, C_blur_jpeg: int64, D_erasing: int64>
child 0, clean: int64
child 1, A_color_geometry: int64
child 2, B_noise: int64
child 3, C_blur_jpeg: int64
child 4, D_erasing: int64
total_frames: int64
clean_frame_fraction: double
clean_episode_fraction: double
total_episodes: int64
sampling: string
frames: struct<clean: int64, A_color_geometry: int64, B_noise: int64, C_blur_jpeg: int64, D_erasing: int64>
child 0, clean: int64
child 1, A_color_geometry: int64
child 2, B_noise: int64
child 3, C_blur_jpeg: int64
child 4, D_erasing: int64
to
{'episodes': {'clean': Value('int64'), 'A_color_geometry': Value('int64'), 'B_noise': Value('int64'), 'C_blur_jpeg': Value('int64'), 'D_erasing': Value('int64')}, 'frames': {'clean': Value('int64'), 'A_color_geometry': Value('int64'), 'B_noise': Value('int64'), 'C_blur_jpeg': Value('int64'), 'D_erasing': Value('int64')}, 'total_episodes': Value('int64'), 'total_frames': Value('int64'), 'clean_episode_fraction': Value('float64'), 'clean_frame_fraction': Value('float64'), 'sampling': Value('string')}
because column names don't match
Traceback: 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 343, in _generate_tables
self._cast_table(pa_table, json_field_paths=json_field_paths),
~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
pa_table = table_cast(pa_table, self.info.features.arrow_schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2378, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2306, in cast_table_to_schema
raise CastError(
...<3 lines>...
)
datasets.table.CastError: Couldn't cast
norm_stats: struct<action.arm.position: struct<min: list<item: double>, max: list<item: double>, mean: list<item (... 1062 chars omitted)
child 0, action.arm.position: struct<min: list<item: double>, max: list<item: double>, mean: list<item: double>, std: list<item: d (... 158 chars omitted)
child 0, min: list<item: double>
child 0, item: double
child 1, max: list<item: double>
child 0, item: double
child 2, mean: list<item: double>
child 0, item: double
child 3, std: list<item: double>
child 0, item: double
child 4, count: list<item: int64>
child 0, item: int64
child 5, q01: list<item: double>
child 0, item: double
child 6, q10: list<item: double>
child 0, item: double
child 7, q50: list<item: double>
child 0, item: double
child 8, q90: list<item: double>
child 0, item: double
child 9, q99: list<item: double>
child 0, item: double
child 1, action.effector.position: struct<min: list<item: double>, max: list<item: double>, mean: list<item: double>, std: list<item: d (... 158 chars omitted)
child 0, min: list<item: double>
child 0, item: double
child 1, max: list<item: double>
child 0, item: double
child 2, mean: list<item: double>
child 0, item: double
child 3, std: list<item: double>
child 0, item: double
child 4, count: list<item: int64>
...
d 3, observation.state.effector.position: struct<min: list<item: double>, max: list<item: double>, mean: list<item: double>, std: list<item: d (... 158 chars omitted)
child 0, min: list<item: double>
child 0, item: double
child 1, max: list<item: double>
child 0, item: double
child 2, mean: list<item: double>
child 0, item: double
child 3, std: list<item: double>
child 0, item: double
child 4, count: list<item: int64>
child 0, item: int64
child 5, q01: list<item: double>
child 0, item: double
child 6, q10: list<item: double>
child 0, item: double
child 7, q50: list<item: double>
child 0, item: double
child 8, q90: list<item: double>
child 0, item: double
child 9, q99: list<item: double>
child 0, item: double
episodes: struct<clean: int64, A_color_geometry: int64, B_noise: int64, C_blur_jpeg: int64, D_erasing: int64>
child 0, clean: int64
child 1, A_color_geometry: int64
child 2, B_noise: int64
child 3, C_blur_jpeg: int64
child 4, D_erasing: int64
total_frames: int64
clean_frame_fraction: double
clean_episode_fraction: double
total_episodes: int64
sampling: string
frames: struct<clean: int64, A_color_geometry: int64, B_noise: int64, C_blur_jpeg: int64, D_erasing: int64>
child 0, clean: int64
child 1, A_color_geometry: int64
child 2, B_noise: int64
child 3, C_blur_jpeg: int64
child 4, D_erasing: int64
to
{'episodes': {'clean': Value('int64'), 'A_color_geometry': Value('int64'), 'B_noise': Value('int64'), 'C_blur_jpeg': Value('int64'), 'D_erasing': Value('int64')}, 'frames': {'clean': Value('int64'), 'A_color_geometry': Value('int64'), 'B_noise': Value('int64'), 'C_blur_jpeg': Value('int64'), 'D_erasing': Value('int64')}, 'total_episodes': Value('int64'), 'total_frames': Value('int64'), 'clean_episode_fraction': Value('float64'), 'clean_frame_fraction': Value('float64'), 'sampling': Value('string')}
because column names don't matchNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
LingBot reviewed clean + offline augmentation mixture
50 per-task LeRobot v3 datasets, combining 2375 reviewed clean and 4250 augmented episodes (6625 total). Original raw files and old caches are not modified.
Layout: datasets/TASK-clean-aug85/{meta,data,videos,provenance}, instructions/TASK/aloha-agilex_clean_50/instructions/episodeN.json, training/{train_mixed.txt,robotwin_clean_aug85_mixed.yaml,norm_stats.json,robot_configs/robotwin.yaml,mix_summary.json}.
Within each task, reviewed clean episodes come first, then A/B/C/D augmented episodes. provenance/episodeN.json maps every new index back to its source. Only seen language pools are retained. No unseen text is used for training. Each source of N observations becomes N-1 training rows: image/state at t, absolute joint target q[t+1]. This follows RoboTwin policy/pi0/scripts/process_data.py. Final observation is omitted, never paired with a fabricated action. This does not switch the model to delta actions. Robot configuration stays absolute.
Videos preserve native 320x240 resolution using browser-compatible libx264, yuv420p, CRF 16, fast preset, GOP 10, no B frames, MP4 faststart. Color tags are explicit BT.601/SMPTE170M limited-range YUV. This is NOT lossless: chroma is subsampled and compressed. Each video is rebuilt directly from the source HDF5, not from a previously lossy video. The original RGB/HDF5 data remains untouched. Decoder FPS is 50, timestamps are verified for every frame. First/middle/last decoded RGB frames must have per-channel MAE<10 and abs channel mean bias<5 vs source; actual metrics are saved in each episode's provenance. LingBot/TorchCodec also checks colors against the source, not just tensor shapes. Parquet actions and states are checked against every source row. Numeric stats and quantiles are recomputed from the actual mixture. Image statistics use uniformly sampled decoded video frames/pixels; RGB is handled by the model image processor.
training/robotwin_clean_aug85_mixed.yaml is derived from the existing robotwin_clean_50_aug_3xa6000.yaml. Only data paths, offline augmentation handling, output directory and resume flag are changed. Existing model/checkpoint/compute hyperparameters are inherited, NOT re-selected or validated for a new machine. Run from /mnt/disk/wyb/LingBot/lingbot-vla-v2 using the established training entry point. Do not launch until status.json says complete and loader_validation.json reports pass. No training is launched by conversion.
Mixing is by concatenating all frames once: no forced 10% clean, no extra group weights or task balancing changes. Actual proportions are in mix_summary.json. Online image augmentation is disabled; seen instruction sampling stays enabled. Training requires this output's instruction_root and robot_config_root; do not reuse the old source episode indices, audit exclusion indices, or old stats.
Visual previews: ../robotwin_clean_offline_aug85_v1/visualizations/index.html The HTML and its sibling PNG/MP4 files are directly viewable locally after copying the whole visualizations folder. These lossy preview MP4s are NOT training files.
- Downloads last month
- 190