Dataset Viewer
The dataset viewer is not available for this split.
Cannot load the dataset split (in streaming mode) to extract the first rows.
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
joint_positions: list<item: double>
child 0, item: double
joint_velocities: list<item: double>
child 0, item: double
effort: list<item: double>
child 0, item: double
ee_positions: list<item: double>
child 0, item: double
gripper_width: double
gripper_velocity: double
timestamp: double
frames: int64
end_time: double
start_time: double
features: struct<cam_global: struct<intrinsics: list<item: list<item: double>>, extrinsics: struct<arms: struc (... 243 chars omitted)
child 0, cam_global: struct<intrinsics: list<item: list<item: double>>, extrinsics: struct<arms: struct<arm: list<item: l (... 21 chars omitted)
child 0, intrinsics: list<item: list<item: double>>
child 0, item: list<item: double>
child 0, item: double
child 1, extrinsics: struct<arms: struct<arm: list<item: list<item: double>>>>
child 0, arms: struct<arm: list<item: list<item: double>>>
child 0, arm: list<item: list<item: double>>
child 0, item: list<item: double>
child 0, item: double
child 1, cam_side: struct<intrinsics: list<item: list<item: double>>, extrinsics: struct<arms: struct<arm: list<item: l (... 21 chars omitted)
child 0, intrinsics: list<item: list<item: double>>
child 0, item: list<item: double>
child 0, item: double
child 1, extrinsics: struct<arms: struct<arm: list<item: list<item: double>>>>
child 0, arms: struct<arm: list<item: list<item: double>>>
child 0, arm: list<item: list<item: double>>
child 0, item: list<item: double>
child 0, item: double
child 2, cam_arm: struct<intrinsics: list<item: null>, extrinsics: struct<>>
child 0, intrinsics: list<item: null>
child 0, item: null
child 1, extrinsics: struct<>
robot_id: string
to
{'start_time': Value('float64'), 'end_time': Value('float64'), 'frames': Value('int64'), 'robot_id': Value('string'), 'features': {'cam_global': {'intrinsics': List(List(Value('float64'))), 'extrinsics': {'arms': {'arm': List(List(Value('float64')))}}}, 'cam_side': {'intrinsics': List(List(Value('float64'))), 'extrinsics': {'arms': {'arm': List(List(Value('float64')))}}}, 'cam_arm': {'intrinsics': List(Value('null')), 'extrinsics': {}}}}
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 478, 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 2818, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2355, in __iter__
for key, pa_table in self._iter_arrow():
~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2380, 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 536, in _iter_arrow
for key, pa_table in iterator:
^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 419, 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 2369, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2297, in cast_table_to_schema
raise CastError(
...<3 lines>...
)
datasets.table.CastError: Couldn't cast
joint_positions: list<item: double>
child 0, item: double
joint_velocities: list<item: double>
child 0, item: double
effort: list<item: double>
child 0, item: double
ee_positions: list<item: double>
child 0, item: double
gripper_width: double
gripper_velocity: double
timestamp: double
frames: int64
end_time: double
start_time: double
features: struct<cam_global: struct<intrinsics: list<item: list<item: double>>, extrinsics: struct<arms: struc (... 243 chars omitted)
child 0, cam_global: struct<intrinsics: list<item: list<item: double>>, extrinsics: struct<arms: struct<arm: list<item: l (... 21 chars omitted)
child 0, intrinsics: list<item: list<item: double>>
child 0, item: list<item: double>
child 0, item: double
child 1, extrinsics: struct<arms: struct<arm: list<item: list<item: double>>>>
child 0, arms: struct<arm: list<item: list<item: double>>>
child 0, arm: list<item: list<item: double>>
child 0, item: list<item: double>
child 0, item: double
child 1, cam_side: struct<intrinsics: list<item: list<item: double>>, extrinsics: struct<arms: struct<arm: list<item: l (... 21 chars omitted)
child 0, intrinsics: list<item: list<item: double>>
child 0, item: list<item: double>
child 0, item: double
child 1, extrinsics: struct<arms: struct<arm: list<item: list<item: double>>>>
child 0, arms: struct<arm: list<item: list<item: double>>>
child 0, arm: list<item: list<item: double>>
child 0, item: list<item: double>
child 0, item: double
child 2, cam_arm: struct<intrinsics: list<item: null>, extrinsics: struct<>>
child 0, intrinsics: list<item: null>
child 0, item: null
child 1, extrinsics: struct<>
robot_id: string
to
{'start_time': Value('float64'), 'end_time': Value('float64'), 'frames': Value('int64'), 'robot_id': Value('string'), 'features': {'cam_global': {'intrinsics': List(List(Value('float64'))), 'extrinsics': {'arms': {'arm': List(List(Value('float64')))}}}, 'cam_side': {'intrinsics': List(List(Value('float64'))), 'extrinsics': {'arms': {'arm': List(List(Value('float64')))}}}, 'cam_arm': {'intrinsics': List(Value('null')), 'extrinsics': {}}}}
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.
Table30v2-Arx5 — press_the_button
Single real-robot manipulation task from RoboChallenge Table30v2, captured on a
single-arm ARX5 (robot_id: rc_arx5_5). One task per repository — intended for
per-task (specialist) training of VLA policies such as DM0 in
Dexbotic (no dataset concatenation).
Task instruction
Press the buttons in the following sequence: pink, blue, green, and then yellow.
Sensors & action space
| Item | Value |
|---|---|
| Cameras (3) | cam_global, cam_arm, cam_side (RGB mp4, libx264) |
| Native video FPS | 30 |
| State / action | 7-dim end-effector [x, y, z, roll, pitch, yaw, gripper] |
| Gripper (index 6) | absolute (not delta) → non_delta_mask=[6] |
| Euler angles (3,4,5) | stay near 0 → no periodic wrapping |
| Episodes | 1060 |
Per-episode meta/episode_meta.json carries camera intrinsics + extrinsics;
states/states.jsonl is the raw capture. The action is derived online by DM0's
AddAction from state (the training jsonl carries only state).
Frame-rate variants (same videos, different sampling)
All variants share the identical mp4 videos; only the jsonl differs by the frame
stride chosen at conversion. frame_idx keeps the original 30 fps index
(15 fps → 0,2,4,…; 10 fps → 0,3,6,…), so the same videos are simply seeked sparsely.
| Folder | FPS | Stride | Frames (= training samples) | Note |
|---|---|---|---|---|
dexdata/ |
30 | STEP=1 | 908366 | every frame |
dexdata_15fps/ |
15 | STEP=2 | 454453 | intermediate |
dexdata_10fps/ |
10 | STEP=3 | 303144 | matches RoboChallenge 10 Hz control (recommended) |
The official Dexbotic-RoboChallengeInference executes actions at 10 Hz (
duration: 0.1s per step), so 10 fps training is the frame rate aligned with deployment.action_horizon(15/25/35…) is the number of action steps consumed per inference cycle — a chunk length, not a frame rate.
Training set size (recommended 10 fps)
DM0 uses one training sample per jsonl line (chunk_size=50 action steps built from the next 50 consecutive frames). At 10 fps this task has:
| Metric | Value |
|---|---|
| Training samples (10 fps frames) | 303,144 |
| Steps / epoch @ effective batch 8 (bs4 × ga2, 1 GPU) | ~37,893 |
| Suggested ~3 epochs | ~113,679 steps |
(For a quick single-GPU smoke test, 2,000–3,000 steps is enough to validate the pipeline and see the loss drop.)
Repository layout
Table30v2-Arx5-press_the_button/
├── data/
│ └── episode_XXXXXX/
│ ├── videos/{cam_global,cam_arm,cam_side}_rgb.mp4
│ ├── meta/episode_meta.json # frames, camera intrinsics/extrinsics
│ └── states/states.jsonl # raw capture (one line per frame)
├── dexdata/*.jsonl # 30 fps DexData annotations
├── dexdata_15fps/*.jsonl # 15 fps
└── dexdata_10fps/*.jsonl # 10 fps (recommended)
DexData jsonl schema (one line = one frame)
{"images_1": {"type": "video", "url": "episode_000140/videos/cam_global_rgb.mp4", "frame_idx": 0},
"images_2": {"type": "video", "url": "episode_000140/videos/cam_arm_rgb.mp4", "frame_idx": 0},
"images_3": {"type": "video", "url": "episode_000140/videos/cam_side_rgb.mp4", "frame_idx": 0},
"state": [x, y, z, roll, pitch, yaw, gripper],
"prompt": "Press the buttons in the following sequence: pink, blue, green, and then yellow.",
"is_robot": true}
images_*.url is relative to data/, so the two trees must be co-located; point
data_path_prefix at <repo>/data and the annotations at the chosen dexdata* folder.
Usage with Dexbotic
Registered in dexbotic/data/data_source/table30.py as
table30v2_arx5_press_the_button (30 fps) and table30v2_arx5_press_the_button_15fps / _10fps.
hf download shigengtian/Table30v2-Arx5-press_the_button --repo-type dataset \
--local-dir /workspace/press_the_button
# 10 fps (matches 10 Hz control, default), single-task DM0 training
torchrun --nproc_per_node=1 playground/table30_dm0.py --dataset press_the_button
# pick another rate
python playground/table30_dm0.py --dataset press_the_button --fps 15
Provenance & license
Derived from RoboChallenge Table30v2 recordings (single-arm ARX5). Converted to the
DexData format via script/convert_data/. License follows the upstream RoboChallenge
terms — verify before redistribution.
- Downloads last month
- -