The dataset viewer is not available for this subset.
Exception: SplitsNotFoundError
Message: The split names could not be parsed from the dataset config.
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 286, in get_dataset_config_info
for split_generator in builder._split_generators(
~~~~~~~~~~~~~~~~~~~~~~~~~^
StreamingDownloadManager(base_path=builder.base_path, download_config=download_config)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/parquet/parquet.py", line 127, in _split_generators
self.info.features = datasets.Features.from_arrow_schema(pq.read_schema(f))
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/features/features.py", line 1971, in from_arrow_schema
metadata_features = Features.from_dict(metadata["info"]["features"])
File "/usr/local/lib/python3.14/site-packages/datasets/features/features.py", line 2009, in from_dict
obj = generate_from_dict(dic)
File "/usr/local/lib/python3.14/site-packages/datasets/features/features.py", line 1588, in generate_from_dict
return {key: generate_from_dict(value) for key, value in obj.items()}
~~~~~~~~~~~~~~~~~~^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/features/features.py", line 1588, in generate_from_dict
return {key: generate_from_dict(value) for key, value in obj.items()}
~~~~~~~~~~~~~~~~~~^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/features/features.py", line 1588, in generate_from_dict
return {key: generate_from_dict(value) for key, value in obj.items()}
^^^^^^^^^
AttributeError: 'str' object has no attribute 'items'
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/split_names.py", line 68, in compute_split_names_from_streaming_response
for split in get_dataset_split_names(
~~~~~~~~~~~~~~~~~~~~~~~^
path=dataset,
^^^^^^^^^^^^^
config_name=config,
^^^^^^^^^^^^^^^^^^^
token=hf_token,
^^^^^^^^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 340, in get_dataset_split_names
info = get_dataset_config_info(
path,
...<6 lines>...
**config_kwargs,
)
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 291, in get_dataset_config_info
raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
datasets.inspect.SplitsNotFoundError: The split names could not be parsed from the dataset config.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.
ctr-scan-object-concurrent-20260916
Complete paired corrected release
Task: scan_object; timing: concurrent. 50 successful episodes, 11251 frames, 25 FPS, LeRobot v3.0.
The collection now contains all 750 successful trajectories. Previously missing 21 trajectories were repaired by restoring the original gripper initialization and scene settling sequence before frozen-control replay. All original seed sets, object identities, control hashes and timing offsets were preserved; no replanning or relaxed collision/success gates. Each task has 50 concurrent, 50 left-first, 50 right-first and 100 uniform episodes using the same original 50 seeds across timing variants.
Ray tracing with OIDN2.5.1, 32 samples/pixel, path depth8; centered_fovy90 wrists at 320x240 and near=0.005m. Measured gripper state is recorded separately from commands. This revision supersedes the successful-only incomplete release; historical revisions remain available. Lifecycle: reported, audit pending.
Included seeds: 5, 7, 9, 12, 26, 30, 34, 58, 59, 71, 75, 76, 78, 82, 84, 86, 87, 90, 95, 98, 107, 118, 119, 127, 130, 135, 138, 141, 146, 150, 151, 152, 154, 155, 159, 160, 164, 170, 172, 174, 185, 187, 189, 190, 191, 200, 201, 202, 205, 211
Joint dimensions and units
Original feature names and their order are preserved exactly in both action and observation.state. Each name below identifies one scalar dimension. Observation values and requested action targets are different quantities.
Left arm
| Index | Original joint name | Unit |
|---|---|---|
| 0 | left_joint_1.pos |
rad |
| 1 | left_joint_2.pos |
rad |
| 2 | left_joint_3.pos |
rad |
| 3 | left_joint_4.pos |
rad |
| 4 | left_joint_5.pos |
rad |
| 5 | left_joint_6.pos |
rad |
| 6 | left_gripper.pos |
0–1 (closed–open) |
Right arm
| Index | Original joint name | Unit |
|---|---|---|
| 7 | right_joint_1.pos |
rad |
| 8 | right_joint_2.pos |
rad |
| 9 | right_joint_3.pos |
rad |
| 10 | right_joint_4.pos |
rad |
| 11 | right_joint_5.pos |
rad |
| 12 | right_joint_6.pos |
rad |
| 13 | right_gripper.pos |
0–1 (closed–open) |
The earlier metadata-only update preserved all data values. The gripper correction documented below changes the confirmed gripper action labels and their statistics; original feature names, observation states, arm actions, videos and timing remain unchanged.
Gripper action correction — 2026-09-16
This revision corrects 6,220 stale initial-hold gripper labels in this dataset. Before each arm's first gripper command, the requested target is held at 1 (open), as established by initialization and the frozen source program. Previous affected labels were 0 (closed).
Reconstruction follows the actual per-physics-step source indices, including
shoe workspace waits and the shared scanning tail. All later gripper commands
match their frozen source values. The per-episode evidence is in
meta/gripper-label-repair.json. Action statistics are recomputed.
Measured observation states, arm action values, original joint names, images, timestamps, masks and episode order are unchanged. Requested targets are not substitutes for measured gripper positions; contact/tracking differences in measured states are preserved. This is a verified action-label correction, not a claim that all historical physical-state provenance has been re-audited.
Supersedes dataset revision 90f7a6e101a697241c95cd79f3223a26097374cb. Old revisions remain accessible.
Old checkpoints require retraining on corrected labels and new normalization
before being used as corrected-data evidence. No training was started here.
Result reported; archive audit approval remains pending.
Repository name
The date suffix is the corrected-release date, 2026-09-16. This repository was renamed from Shiki42/ctr-scan-object-concurrent-20260911 to Shiki42/ctr-scan-object-concurrent-20260916; dataset payloads and original joint names are unchanged. Historical commits remain available.
Arm active mask
observation.arm_active_mask was reconstructed on 2026-09-17 from the pinned
meta/retime_manifest.json and meta/retime_source_indices/*.npz traces. It is
float32[2], ordered left/right. At observation t, value 1 means that arm's
actual next action interval contains task-program progress, including gripper
program progress. All waits and holds are 0. No velocity threshold is
used and no synthetic terminal frame is added. retime.*_idle is a
distinct loss-mask annotation; necessary waits can remain un-masked for
loss while still being inactive here. Per-episode evidence is in
meta/arm-active-mask-contract.json.
The replacement names and value semantics match the existing
Shiki42/PutCab-CTR-Train50-V4 field.
- Downloads last month
- 493