The dataset viewer is not available for this split.
Error code: StreamingRowsError
Exception: TypeError
Message: Couldn't cast array of type
struct<beat_block_hammer-demo_clean_collect_200-50: struct<0: int64, 1: int64, 2: int64, 3: int64, 4: int64, 5: int64, 6: int64, 7: int64, 8: int64, 9: int64, 10: int64, 11: int64, 12: int64, 13: int64, 14: int64, 15: int64, 16: int64, 17: int64, 18: int64, 19: int64, 20: int64, 21: int64, 22: int64, 23: int64, 24: int64, 25: int64, 26: int64, 27: int64, 28: int64, 29: int64, 30: int64, 31: int64, 32: int64, 33: int64, 34: int64, 35: int64, 36: int64, 37: int64, 38: int64, 39: int64, 40: int64, 41: int64, 42: int64, 43: int64, 44: int64, 45: int64, 46: int64, 47: int64, 48: int64, 49: int64>>
to
{'adjust_bottle-demo_clean_collect_200-50': {'0': Value('int64'), '1': Value('int64'), '2': Value('int64'), '3': Value('int64'), '4': Value('int64'), '5': Value('int64'), '6': Value('int64'), '7': Value('int64'), '8': Value('int64'), '9': Value('int64'), '10': Value('int64'), '11': Value('int64'), '12': Value('int64'), '13': Value('int64'), '14': Value('int64'), '15': Value('int64'), '16': Value('int64'), '17': Value('int64'), '18': Value('int64'), '19': Value('int64'), '20': Value('int64'), '21': Value('int64'), '22': Value('int64'), '23': Value('int64'), '24': Value('int64'), '25': Value('int64'), '26': Value('int64'), '27': Value('int64'), '28': Value('int64'), '29': Value('int64'), '30': Value('int64'), '31': Value('int64'), '32': Value('int64'), '33': Value('int64'), '34': Value('int64'), '35': Value('int64'), '36': Value('int64'), '37': Value('int64'), '38': Value('int64'), '39': Value('int64'), '40': Value('int64'), '41': Value('int64'), '42': Value('int64'), '43': Value('int64'), '44': Value('int64'), '45': Value('int64'), '46': Value('int64'), '47': Value('int64'), '48': Value('int64'), '49': Value('int64')}}
Traceback: Traceback (most recent call last):
File "/src/services/worker/src/worker/utils.py", line 149, 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 274, in decorator
return func(*args, **kwargs)
File "/src/services/worker/src/worker/utils.py", line 129, 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 544, 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 3074, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2578, in __iter__
for key, pa_table in self._iter_arrow():
~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2603, 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 577, in _iter_arrow
for key, pa_table in iterator:
^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 459, 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 373, 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 2394, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2328, in cast_table_to_schema
cast_array_to_feature(
~~~~~~~~~~~~~~~~~~~~~^
table[name] if name in table_column_names else pa.array([None] * len(table), type=schema.field(name).type),
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
feature,
^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 1870, in wrapper
return pa.chunked_array([func(chunk, *args, **kwargs) for chunk in array.chunks])
~~~~^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2080, in cast_array_to_feature
_c(children[name] if name in array_fields else null_array, subfeature)
~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 1872, in wrapper
return func(array, *args, **kwargs)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2170, in cast_array_to_feature
raise TypeError(f"Couldn't cast array of type\n{_short_str(array.type)}\nto\n{_short_str(feature)}")
TypeError: Couldn't cast array of type
struct<beat_block_hammer-demo_clean_collect_200-50: struct<0: int64, 1: int64, 2: int64, 3: int64, 4: int64, 5: int64, 6: int64, 7: int64, 8: int64, 9: int64, 10: int64, 11: int64, 12: int64, 13: int64, 14: int64, 15: int64, 16: int64, 17: int64, 18: int64, 19: int64, 20: int64, 21: int64, 22: int64, 23: int64, 24: int64, 25: int64, 26: int64, 27: int64, 28: int64, 29: int64, 30: int64, 31: int64, 32: int64, 33: int64, 34: int64, 35: int64, 36: int64, 37: int64, 38: int64, 39: int64, 40: int64, 41: int64, 42: int64, 43: int64, 44: int64, 45: int64, 46: int64, 47: int64, 48: int64, 49: int64>>
to
{'adjust_bottle-demo_clean_collect_200-50': {'0': Value('int64'), '1': Value('int64'), '2': Value('int64'), '3': Value('int64'), '4': Value('int64'), '5': Value('int64'), '6': Value('int64'), '7': Value('int64'), '8': Value('int64'), '9': Value('int64'), '10': Value('int64'), '11': Value('int64'), '12': Value('int64'), '13': Value('int64'), '14': Value('int64'), '15': Value('int64'), '16': Value('int64'), '17': Value('int64'), '18': Value('int64'), '19': Value('int64'), '20': Value('int64'), '21': Value('int64'), '22': Value('int64'), '23': Value('int64'), '24': Value('int64'), '25': Value('int64'), '26': Value('int64'), '27': Value('int64'), '28': Value('int64'), '29': Value('int64'), '30': Value('int64'), '31': Value('int64'), '32': Value('int64'), '33': Value('int64'), '34': Value('int64'), '35': Value('int64'), '36': Value('int64'), '37': Value('int64'), '38': Value('int64'), '39': Value('int64'), '40': Value('int64'), '41': Value('int64'), '42': Value('int64'), '43': Value('int64'), '44': Value('int64'), '45': Value('int64'), '46': Value('int64'), '47': Value('int64'), '48': Value('int64'), '49': Value('int64')}}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.
RoboTwin Clean MoSE3 supervision cache
Offline MoSE3 supervision for the FastWAM-Joint and LingBot-VA RoboTwin Clean→Random experiments. Both training pipelines use the same source windows and camera order. The 50 task directories hold 2,527 Safetensors shards and 50 training indexes for 30,551 windows from 2,492 episodes.
The source is the EEF Clean split at revision 595d4acc7e026aa57afba218f4170c6d79e39103 of the RoboTwin LeRobot release. The upstream catalogue has 2,500 episodes. Eight lack one released wrist VAE file and were excluded from both experiments. Their identifiers, the teacher checkpoint hash, the task counts and SHA-256 checksums of all cache files are in the manifest.
Which VAE input does each RoboTwin trainer use?
Both backbones read the same 30,551 MoSE3 supervision windows from this repository. Set LINGBOT_MOSE3_CACHE to its local root and LINGBOT_CLEAN_DATA to the upstream lerobot_robotwin_eef_clean_50 directory. Keep each task’s meta/, data/, videos/ and latents/ together. Official EEF Clean split
Run the corresponding command from the root of a CoMove checkout. The launcher links require access to the CoMove GitHub repository.
| Training backbone | Video input | Training launcher |
|---|---|---|
| FastWAM-Joint | The loader tiles head RGB above the two wrists on a 384 × 320 canvas (height × width). The frozen Wan2.2 VAE encodes each nine-frame window online. The published camera frame IDs align the windows. | scripts/train/robotwin_clean2random/fastwam_joint_online_vae.sh 8 |
| LingBot-VA | The loader reads the head, left-wrist and right-wrist VAE tensors from each task’s latents/chunk-*/observation.images.cam_*/ files and joins their grids with the wrists above the head. |
scripts/train/robotwin_clean2random/lingbot_va_cached_vae.sh 8 |
The two VAE paths produce different video features. In a check on 48 matched windows from 16 episodes in eight tasks, zero pairs were elementwise identical after aligning the camera layout. Both trainers can reuse the same MoSE3 supervision for those frames. The two-camera, 224-pixel FastWAM VAE cache belongs to the separate LIBERO experiment. LIBERO VAE dataset
Read the cache
Download this dataset and set LINGBOT_MOSE3_CACHE to its root. Each task has an index.json mapping <task>/<episode>/<start> to a shard filename and row. The original release supplies the actions and each backbone’s video inputs as listed above.
Each complete window spans 33 control frames. The camera order is head, left wrist, right wrist. The eight future offsets are 4, 8, 12, 16, 20, 24, 28 and 32 control steps from the anchor. Each camera and offset has 2,048 points on a 224 × 224 image-coordinate grid and a 16-dimensional teacher embedding per point. Shards contain embedding, pixel_yx, point_valid, horizon_valid and observed_length.
All 30,551 keys were checked against the upstream VAE time grid and both training datasets. Additional checks compared 55 materialized pairs across all tasks, recomputed 10 MoSE3 windows from source video, and ran two forward and backward training steps for each backbone.
- Downloads last month
- -