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
state: struct<proprio: struct<start: int64, end: int64, dtype: string, absolute: bool, original_key: string (... 2 chars omitted)
child 0, proprio: struct<start: int64, end: int64, dtype: string, absolute: bool, original_key: string>
child 0, start: int64
child 1, end: int64
child 2, dtype: string
child 3, absolute: bool
child 4, original_key: string
action: struct<eef: struct<start: int64, end: int64, dtype: string, absolute: bool, original_key: string>>
child 0, eef: struct<start: int64, end: int64, dtype: string, absolute: bool, original_key: string>
child 0, start: int64
child 1, end: int64
child 2, dtype: string
child 3, absolute: bool
child 4, original_key: string
video: struct<top: struct<original_key: string>, wrist: struct<original_key: string>>
child 0, top: struct<original_key: string>
child 0, original_key: string
child 1, wrist: struct<original_key: string>
child 0, original_key: string
annotation: struct<human.task_description: struct<original_key: string>>
child 0, human.task_description: struct<original_key: string>
child 0, original_key: string
data_files_size_in_mb: int64
total_tasks: int64
features: struct<observation.images.top: struct<dtype: string, shape: list<item: int64>, names: list<item: str (... 967 chars omitted)
child 0, observation.images.top: struct<dtype: string, shape: list<item: int64>, names: list<item: string>, info: struct<video.height (... 157 chars
...
odec: string
child 3, video.pix_fmt: string
child 4, video.is_depth_map: bool
child 5, video.fps: int64
child 6, video.channels: int64
child 7, has_audio: bool
child 4, timestamp: struct<dtype: string, shape: list<item: int64>, names: null>
child 0, dtype: string
child 1, shape: list<item: int64>
child 0, item: int64
child 2, names: null
child 5, frame_index: struct<dtype: string, shape: list<item: int64>, names: null>
child 0, dtype: string
child 1, shape: list<item: int64>
child 0, item: int64
child 2, names: null
child 6, episode_index: struct<dtype: string, shape: list<item: int64>, names: null>
child 0, dtype: string
child 1, shape: list<item: int64>
child 0, item: int64
child 2, names: null
child 7, index: struct<dtype: string, shape: list<item: int64>, names: null>
child 0, dtype: string
child 1, shape: list<item: int64>
child 0, item: int64
child 2, names: null
child 8, task_index: struct<dtype: string, shape: list<item: int64>, names: null>
child 0, dtype: string
child 1, shape: list<item: int64>
child 0, item: int64
child 2, names: null
total_frames: int64
codebase_version: string
total_episodes: int64
fps: int64
video_path: string
data_path: string
chunks_size: int64
robot_type: string
splits: struct<train: string>
child 0, train: string
video_files_size_in_mb: int64
to
{'codebase_version': Value('string'), 'robot_type': Value('string'), 'total_episodes': Value('int64'), 'total_frames': Value('int64'), 'total_tasks': Value('int64'), 'chunks_size': Value('int64'), 'data_files_size_in_mb': Value('int64'), 'video_files_size_in_mb': Value('int64'), 'fps': Value('int64'), 'splits': {'train': Value('string')}, 'data_path': Value('string'), 'video_path': Value('string'), 'features': {'observation.images.top': {'dtype': Value('string'), 'shape': List(Value('int64')), 'names': List(Value('string')), 'info': {'video.height': Value('int64'), 'video.width': Value('int64'), 'video.codec': Value('string'), 'video.pix_fmt': Value('string'), 'video.is_depth_map': Value('bool'), 'video.fps': Value('int64'), 'video.channels': Value('int64'), 'has_audio': Value('bool')}}, 'observation.state': {'dtype': Value('string'), 'shape': List(Value('int64'))}, 'action': {'dtype': Value('string'), 'shape': List(Value('int64'))}, 'observation.images.wrist': {'dtype': Value('string'), 'shape': List(Value('int64')), 'names': List(Value('string')), 'info': {'video.height': Value('int64'), 'video.width': Value('int64'), 'video.codec': Value('string'), 'video.pix_fmt': Value('string'), 'video.is_depth_map': Value('bool'), 'video.fps': Value('int64'), 'video.channels': Value('int64'), 'has_audio': Value('bool')}}, 'timestamp': {'dtype': Value('string'), 'shape': List(Value('int64')), 'names': Value('null')}, 'frame_index': {'dtype': Value('string'), 'shape': List(Value('int64')), 'names': Value('null')}, 'episode_index': {'dtype': Value('string'), 'shape': List(Value('int64')), 'names': Value('null')}, 'index': {'dtype': Value('string'), 'shape': List(Value('int64')), 'names': Value('null')}, 'task_index': {'dtype': Value('string'), 'shape': List(Value('int64')), 'names': Value('null')}}}
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 2840, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2373, in __iter__
for key, pa_table in self._iter_arrow():
~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2398, 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 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
state: struct<proprio: struct<start: int64, end: int64, dtype: string, absolute: bool, original_key: string (... 2 chars omitted)
child 0, proprio: struct<start: int64, end: int64, dtype: string, absolute: bool, original_key: string>
child 0, start: int64
child 1, end: int64
child 2, dtype: string
child 3, absolute: bool
child 4, original_key: string
action: struct<eef: struct<start: int64, end: int64, dtype: string, absolute: bool, original_key: string>>
child 0, eef: struct<start: int64, end: int64, dtype: string, absolute: bool, original_key: string>
child 0, start: int64
child 1, end: int64
child 2, dtype: string
child 3, absolute: bool
child 4, original_key: string
video: struct<top: struct<original_key: string>, wrist: struct<original_key: string>>
child 0, top: struct<original_key: string>
child 0, original_key: string
child 1, wrist: struct<original_key: string>
child 0, original_key: string
annotation: struct<human.task_description: struct<original_key: string>>
child 0, human.task_description: struct<original_key: string>
child 0, original_key: string
data_files_size_in_mb: int64
total_tasks: int64
features: struct<observation.images.top: struct<dtype: string, shape: list<item: int64>, names: list<item: str (... 967 chars omitted)
child 0, observation.images.top: struct<dtype: string, shape: list<item: int64>, names: list<item: string>, info: struct<video.height (... 157 chars
...
odec: string
child 3, video.pix_fmt: string
child 4, video.is_depth_map: bool
child 5, video.fps: int64
child 6, video.channels: int64
child 7, has_audio: bool
child 4, timestamp: struct<dtype: string, shape: list<item: int64>, names: null>
child 0, dtype: string
child 1, shape: list<item: int64>
child 0, item: int64
child 2, names: null
child 5, frame_index: struct<dtype: string, shape: list<item: int64>, names: null>
child 0, dtype: string
child 1, shape: list<item: int64>
child 0, item: int64
child 2, names: null
child 6, episode_index: struct<dtype: string, shape: list<item: int64>, names: null>
child 0, dtype: string
child 1, shape: list<item: int64>
child 0, item: int64
child 2, names: null
child 7, index: struct<dtype: string, shape: list<item: int64>, names: null>
child 0, dtype: string
child 1, shape: list<item: int64>
child 0, item: int64
child 2, names: null
child 8, task_index: struct<dtype: string, shape: list<item: int64>, names: null>
child 0, dtype: string
child 1, shape: list<item: int64>
child 0, item: int64
child 2, names: null
total_frames: int64
codebase_version: string
total_episodes: int64
fps: int64
video_path: string
data_path: string
chunks_size: int64
robot_type: string
splits: struct<train: string>
child 0, train: string
video_files_size_in_mb: int64
to
{'codebase_version': Value('string'), 'robot_type': Value('string'), 'total_episodes': Value('int64'), 'total_frames': Value('int64'), 'total_tasks': Value('int64'), 'chunks_size': Value('int64'), 'data_files_size_in_mb': Value('int64'), 'video_files_size_in_mb': Value('int64'), 'fps': Value('int64'), 'splits': {'train': Value('string')}, 'data_path': Value('string'), 'video_path': Value('string'), 'features': {'observation.images.top': {'dtype': Value('string'), 'shape': List(Value('int64')), 'names': List(Value('string')), 'info': {'video.height': Value('int64'), 'video.width': Value('int64'), 'video.codec': Value('string'), 'video.pix_fmt': Value('string'), 'video.is_depth_map': Value('bool'), 'video.fps': Value('int64'), 'video.channels': Value('int64'), 'has_audio': Value('bool')}}, 'observation.state': {'dtype': Value('string'), 'shape': List(Value('int64'))}, 'action': {'dtype': Value('string'), 'shape': List(Value('int64'))}, 'observation.images.wrist': {'dtype': Value('string'), 'shape': List(Value('int64')), 'names': List(Value('string')), 'info': {'video.height': Value('int64'), 'video.width': Value('int64'), 'video.codec': Value('string'), 'video.pix_fmt': Value('string'), 'video.is_depth_map': Value('bool'), 'video.fps': Value('int64'), 'video.channels': Value('int64'), 'has_audio': Value('bool')}}, 'timestamp': {'dtype': Value('string'), 'shape': List(Value('int64')), 'names': Value('null')}, 'frame_index': {'dtype': Value('string'), 'shape': List(Value('int64')), 'names': Value('null')}, 'episode_index': {'dtype': Value('string'), 'shape': List(Value('int64')), 'names': Value('null')}, 'index': {'dtype': Value('string'), 'shape': List(Value('int64')), 'names': Value('null')}, 'task_index': {'dtype': Value('string'), 'shape': List(Value('int64')), 'names': Value('null')}}}
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.
MIKASA InterceptGrabFast H1 rollouts
Successful deterministic rollouts from the PPO teacher for
InterceptGrabFast-VLA-v0. The language instruction is:
Intercept the rolling ball and grasp it to stop it.
The task uses the H1 embodiment, a 7D pd_ee_delta_pose action, top and wrist
RGB observations, and a native 20 Hz control rate. The full reproduction guide
is the MIKASA cookbook.
Contents
data_npz/intercept_grab_fast_vla_v0/ # 275 raw episodes
data_lerobot/
├── local/intercept_grab_fast_vla_v0_h1_train/ # 250 episodes
├── local/intercept_grab_fast_vla_v0_h1_val/ # 25 episodes
└── mikasa_intercept_grab_fast_h1.json
The raw rollout contains 275 successful episodes, 9,859 transitions, and 275
unique seeds. Episode length has mean/min/max 35.851/24/57. Actions are clipped
to [-1, 1]. The LeRobot v3 conversion contains 8,911 train frames and 948
validation frames at 20 FPS. The split preserves collection order: episodes
0:250 train and 250:275 validation.
The RLDS conversion is an intermediate duplicate and is not published.
Raw NPZ schema
Each train_data_*.npz file stores one episode:
| Field | Type and shape | Meaning |
|---|---|---|
rgb |
uint8[T, 128, 128, 6] |
concatenated top and wrist RGB |
proprio |
float32[T, 7] |
robot proprioception |
action |
float32[T, 7] |
end-effector delta pose action |
reward |
float32[T] |
environment reward |
success |
int32[T] |
per-step success signal |
done |
int32[T] |
per-step termination signal |
language_instruction |
scalar string | task instruction |
success_once |
scalar bool | episode success |
episode_length |
scalar int32 | number of transitions |
episode_seed |
scalar int64 | rollout seed |
LeRobot schema
Each split is self-contained with data/, videos/, and meta/:
| Field | Type and shape | Meaning |
|---|---|---|
observation.images.top |
AV1 video, 128 × 128 × 3 |
top camera RGB |
observation.images.wrist |
AV1 video, 128 × 128 × 3 |
wrist camera RGB |
observation.state |
float32[7] |
absolute proprioception |
action |
float32[7] |
non-absolute end-effector action |
timestamp |
float32[1] |
seconds from episode start |
episode_index |
int64[1] |
episode id |
frame_index |
int64[1] |
frame id within episode |
task_index |
int64[1] |
task id |
Use local/intercept_grab_fast_vla_v0_h1_train for training and keep
local/intercept_grab_fast_vla_v0_h1_val for held-out action-loss evaluation.
The mixture JSON records both exact StarVLA dataset names.
Provenance
The teacher uses a 49D privileged state and deterministic mean actions. It was
trained with normalized dense reward, seed 123, learning rate 1e-4, and eight
PPO update epochs. The selected checkpoint passed 100/100 canonical episodes
before collection. The MIKASA integration revision is
cf4f96f319022f89c9d7cfbd639d19bc10ed44fb on vendor baseline
16634db18bef08128ed79346469c86fc12169aed. The dataset contains only complete
successful trajectories.
This dataset is for research and benchmarking in MIKASA. No license metadata is asserted here.
- Downloads last month
- 72