The dataset viewer is not available for this split.
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
conversion_package_tests: struct<passed: int64, failed: int64>
child 0, passed: int64
child 1, failed: int64
ruff_format: string
ruff_check: string
full_repository_tests: struct<exit_code: int64, failed: int64, errors: int64, cause: string, log: string>
child 0, exit_code: int64
child 1, failed: int64
child 2, errors: int64
child 3, cause: string
child 4, log: string
gpu_simulation_launched: bool
data_validation: string
episodes: int64
wrist_from_canonical: list<item: list<item: double>>
child 0, item: list<item: double>
child 0, item: double
status: string
urdf_sha256: string
warmup_camera_estimates: int64
frames: int64
coordinate_frame: string
rotation_tolerance_rad: double
array_sha256: struct<handover: struct<left_wrist_pose: string, right_wrist_pose: string, camera_ego_pose: string, (... 1048 chars omitted)
child 0, handover: struct<left_wrist_pose: string, right_wrist_pose: string, camera_ego_pose: string, dextraverse_camer (... 266 chars omitted)
child 0, left_wrist_pose: string
child 1, right_wrist_pose: string
child 2, camera_ego_pose: string
child 3, dextraverse_camera_sync_pose: string
child 4, dextraverse_camera_warmup_estimate: string
child 5, camera_main_pose: string
child 6, camera_side_pose: string
child 7, dextraverse_pose_reference_valid: string
child 8, dextraverse_pose_reference_position_error_m: string
child 9, dextraverse_pose_reference_rotation_error_rad: string
child
...
: string, demo: string, frames: int64, reference_mismatch_frames: list<item: i (... 305 chars omitted)
child 0, item: struct<task: string, demo: string, frames: int64, reference_mismatch_frames: list<item: int64>, sens (... 293 chars omitted)
child 0, task: string
child 1, demo: string
child 2, frames: int64
child 3, reference_mismatch_frames: list<item: int64>
child 0, item: int64
child 4, sensor_position_error_max_m: double
child 5, sensor_rotation_error_max_rad: double
child 6, sensor_position_error_after_first_max_m: double
child 7, sensor_rotation_error_after_first_max_rad: double
child 8, final_camera_metadata_position_error_m: double
child 9, camera_step_position_max_m: double
child 10, camera_step_rotation_max_rad: double
summary: struct<sensor_reference_passed_frames: int64, sensor_reference_mismatch_first_frames: int64, sensor_ (... 291 chars omitted)
child 0, sensor_reference_passed_frames: int64
child 1, sensor_reference_mismatch_first_frames: int64
child 2, sensor_position_error_after_first_max_m: double
child 3, sensor_rotation_error_after_first_max_rad: double
child 4, final_camera_metadata_position_error_max_m: double
child 5, camera_capture_to_sync_position_step_max_m: double
child 6, camera_capture_to_sync_rotation_step_max_rad: double
child 7, not_a_camera_exposure_error_bound: bool
position_tolerance_m: double
camera_timing: string
reference_mismatch_frames: int64
to
{'status': Value('string'), 'published': Value('bool'), 'frames': Value('int64'), 'episodes': Value('int64'), 'urdf_sha256': Value('string'), 'position_tolerance_m': Value('float64'), 'rotation_tolerance_rad': Value('float64'), 'reference_mismatch_frames': Value('int64'), 'warmup_camera_estimates': Value('int64'), 'coordinate_frame': Value('string'), 'wrist_from_canonical': List(List(Value('float64'))), 'camera_timing': Value('string'), 'episodes_report': List({'task': Value('string'), 'demo': Value('string'), 'frames': Value('int64'), 'reference_mismatch_frames': List(Value('int64')), 'sensor_position_error_max_m': Value('float64'), 'sensor_rotation_error_max_rad': Value('float64'), 'sensor_position_error_after_first_max_m': Value('float64'), 'sensor_rotation_error_after_first_max_rad': Value('float64'), 'final_camera_metadata_position_error_m': Value('float64'), 'camera_step_position_max_m': Value('float64'), 'camera_step_rotation_max_rad': Value('float64')}), 'array_sha256': {'handover': {'left_wrist_pose': Value('string'), 'right_wrist_pose': Value('string'), 'camera_ego_pose': Value('string'), 'dextraverse_camera_sync_pose': Value('string'), 'dextraverse_camera_warmup_estimate': Value('string'), 'camera_main_pose': Value('string'), 'camera_side_pose': Value('string'), 'dextraverse_pose_reference_valid': Value('string'), 'dextraverse_pose_reference_position_error_m': Value('string'), 'dextraverse_pose_reference_rotation_error_rad': Value('string')}, 'pick_and_place': {'left_wrist_pose': Value('string'), 'right_wrist_pose': Value('string'), 'camera_ego_pose': Value('string'), 'dextraverse_camera_sync_pose': Value('string'), 'dextraverse_camera_warmup_estimate': Value('string'), 'camera_main_pose': Value('string'), 'camera_side_pose': Value('string'), 'dextraverse_pose_reference_valid': Value('string'), 'dextraverse_pose_reference_position_error_m': Value('string'), 'dextraverse_pose_reference_rotation_error_rad': Value('string')}, 'stack_cube': {'left_wrist_pose': Value('string'), 'right_wrist_pose': Value('string'), 'camera_ego_pose': Value('string'), 'dextraverse_camera_sync_pose': Value('string'), 'dextraverse_camera_warmup_estimate': Value('string'), 'camera_main_pose': Value('string'), 'camera_side_pose': Value('string'), 'dextraverse_pose_reference_valid': Value('string'), 'dextraverse_pose_reference_position_error_m': Value('string'), 'dextraverse_pose_reference_rotation_error_rad': Value('string')}}, 'summary': {'sensor_reference_passed_frames': Value('int64'), 'sensor_reference_mismatch_first_frames': Value('int64'), 'sensor_position_error_after_first_max_m': Value('float64'), 'sensor_rotation_error_after_first_max_rad': Value('float64'), 'final_camera_metadata_position_error_max_m': Value('float64'), 'camera_capture_to_sync_position_step_max_m': Value('float64'), 'camera_capture_to_sync_rotation_step_max_rad': Value('float64'), 'not_a_camera_exposure_error_bound': Value('bool')}}
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
conversion_package_tests: struct<passed: int64, failed: int64>
child 0, passed: int64
child 1, failed: int64
ruff_format: string
ruff_check: string
full_repository_tests: struct<exit_code: int64, failed: int64, errors: int64, cause: string, log: string>
child 0, exit_code: int64
child 1, failed: int64
child 2, errors: int64
child 3, cause: string
child 4, log: string
gpu_simulation_launched: bool
data_validation: string
episodes: int64
wrist_from_canonical: list<item: list<item: double>>
child 0, item: list<item: double>
child 0, item: double
status: string
urdf_sha256: string
warmup_camera_estimates: int64
frames: int64
coordinate_frame: string
rotation_tolerance_rad: double
array_sha256: struct<handover: struct<left_wrist_pose: string, right_wrist_pose: string, camera_ego_pose: string, (... 1048 chars omitted)
child 0, handover: struct<left_wrist_pose: string, right_wrist_pose: string, camera_ego_pose: string, dextraverse_camer (... 266 chars omitted)
child 0, left_wrist_pose: string
child 1, right_wrist_pose: string
child 2, camera_ego_pose: string
child 3, dextraverse_camera_sync_pose: string
child 4, dextraverse_camera_warmup_estimate: string
child 5, camera_main_pose: string
child 6, camera_side_pose: string
child 7, dextraverse_pose_reference_valid: string
child 8, dextraverse_pose_reference_position_error_m: string
child 9, dextraverse_pose_reference_rotation_error_rad: string
child
...
: string, demo: string, frames: int64, reference_mismatch_frames: list<item: i (... 305 chars omitted)
child 0, item: struct<task: string, demo: string, frames: int64, reference_mismatch_frames: list<item: int64>, sens (... 293 chars omitted)
child 0, task: string
child 1, demo: string
child 2, frames: int64
child 3, reference_mismatch_frames: list<item: int64>
child 0, item: int64
child 4, sensor_position_error_max_m: double
child 5, sensor_rotation_error_max_rad: double
child 6, sensor_position_error_after_first_max_m: double
child 7, sensor_rotation_error_after_first_max_rad: double
child 8, final_camera_metadata_position_error_m: double
child 9, camera_step_position_max_m: double
child 10, camera_step_rotation_max_rad: double
summary: struct<sensor_reference_passed_frames: int64, sensor_reference_mismatch_first_frames: int64, sensor_ (... 291 chars omitted)
child 0, sensor_reference_passed_frames: int64
child 1, sensor_reference_mismatch_first_frames: int64
child 2, sensor_position_error_after_first_max_m: double
child 3, sensor_rotation_error_after_first_max_rad: double
child 4, final_camera_metadata_position_error_max_m: double
child 5, camera_capture_to_sync_position_step_max_m: double
child 6, camera_capture_to_sync_rotation_step_max_rad: double
child 7, not_a_camera_exposure_error_bound: bool
position_tolerance_m: double
camera_timing: string
reference_mismatch_frames: int64
to
{'status': Value('string'), 'published': Value('bool'), 'frames': Value('int64'), 'episodes': Value('int64'), 'urdf_sha256': Value('string'), 'position_tolerance_m': Value('float64'), 'rotation_tolerance_rad': Value('float64'), 'reference_mismatch_frames': Value('int64'), 'warmup_camera_estimates': Value('int64'), 'coordinate_frame': Value('string'), 'wrist_from_canonical': List(List(Value('float64'))), 'camera_timing': Value('string'), 'episodes_report': List({'task': Value('string'), 'demo': Value('string'), 'frames': Value('int64'), 'reference_mismatch_frames': List(Value('int64')), 'sensor_position_error_max_m': Value('float64'), 'sensor_rotation_error_max_rad': Value('float64'), 'sensor_position_error_after_first_max_m': Value('float64'), 'sensor_rotation_error_after_first_max_rad': Value('float64'), 'final_camera_metadata_position_error_m': Value('float64'), 'camera_step_position_max_m': Value('float64'), 'camera_step_rotation_max_rad': Value('float64')}), 'array_sha256': {'handover': {'left_wrist_pose': Value('string'), 'right_wrist_pose': Value('string'), 'camera_ego_pose': Value('string'), 'dextraverse_camera_sync_pose': Value('string'), 'dextraverse_camera_warmup_estimate': Value('string'), 'camera_main_pose': Value('string'), 'camera_side_pose': Value('string'), 'dextraverse_pose_reference_valid': Value('string'), 'dextraverse_pose_reference_position_error_m': Value('string'), 'dextraverse_pose_reference_rotation_error_rad': Value('string')}, 'pick_and_place': {'left_wrist_pose': Value('string'), 'right_wrist_pose': Value('string'), 'camera_ego_pose': Value('string'), 'dextraverse_camera_sync_pose': Value('string'), 'dextraverse_camera_warmup_estimate': Value('string'), 'camera_main_pose': Value('string'), 'camera_side_pose': Value('string'), 'dextraverse_pose_reference_valid': Value('string'), 'dextraverse_pose_reference_position_error_m': Value('string'), 'dextraverse_pose_reference_rotation_error_rad': Value('string')}, 'stack_cube': {'left_wrist_pose': Value('string'), 'right_wrist_pose': Value('string'), 'camera_ego_pose': Value('string'), 'dextraverse_camera_sync_pose': Value('string'), 'dextraverse_camera_warmup_estimate': Value('string'), 'camera_main_pose': Value('string'), 'camera_side_pose': Value('string'), 'dextraverse_pose_reference_valid': Value('string'), 'dextraverse_pose_reference_position_error_m': Value('string'), 'dextraverse_pose_reference_rotation_error_rad': Value('string')}}, 'summary': {'sensor_reference_passed_frames': Value('int64'), 'sensor_reference_mismatch_first_frames': Value('int64'), 'sensor_position_error_after_first_max_m': Value('float64'), 'sensor_rotation_error_after_first_max_rad': Value('float64'), 'final_camera_metadata_position_error_max_m': Value('float64'), 'camera_capture_to_sync_position_step_max_m': Value('float64'), 'camera_capture_to_sync_rotation_step_max_rad': Value('float64'), 'not_a_camera_exposure_error_bound': Value('bool')}}
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.
DextraVerse for Open-X-Tactile
Three simulated dexterous manipulation tasks with Vega arms and Sharpa hands. RGB is rendered, TacMap deformation images are synthetic, and contact forces come from Isaac PhysX. These are simulation observations, not physical sensor measurements.
Data conventions and the submission format are described by Open-X-Tactile (OXT) and the linked FTP-1 processing specification.
| Archive | Task | Episodes | Frames |
|---|---|---|---|
DextraVerse/handover.zarr.zip |
Handover | 100 | 57,681 |
DextraVerse/pick_and_place.zarr.zip |
Pick and place | 100 | 41,684 |
DextraVerse/stack_cube.zarr.zip |
Stack cube | 100 | 36,369 |
| Total | 3 tasks | 300 | 135,734 |
Each archive contains its own <task>.zarr/ root including hidden Zarr metadata.
The ZIPs use ZIP64 and store already-compressed Zarr chunks without an additional compression pass.
SHA256SUMS.txt records the archive hashes. Keep the DextraVerse/ subdirectory when uploading this repository to Hugging Face or ModelScope.
Reading the data
Extract each archive into a local directory. Python with zarr==2.18.7, numcodecs==0.15.1 and NumPy was used for this package's checks.
import zarr
root = zarr.open_group('handover.zarr', mode='r')
episode_end = int(root['meta/episode_ends'][0])
rgb = root['data/camera_main_rgb'][:episode_end]
joints = root['data/right_hand_joints'][:episode_end]
Every task has 46 time-major arrays in data/ and 100 exclusive episode boundaries in meta/episode_ends.
This revision regroups the same 63 observed joint coordinates into 22 left-hand, 22 right-hand, 7 left-arm, 7 right-arm and 5 supplementary coordinates. Five joint/index arrays and layout metadata were rewritten. All other source members are verified byte for byte against the baseline. No observation value, source sign, unit, zero position, time alignment or episode boundary was changed.
Modalities and conventions
- Three RGB views (
camera_main_rgb,camera_ego_rgb,camera_side_rgb): uint8,(T,224,224,3). - Left/right arm joints:
(T,7), radians. Left/right hand joints:(T,22), radians, paired with canonical indices. - Canonical hand indices in stored column order:
[1,2,3,4,6,7,8,9,11,12,13,14,16,17,18,19,21,22,23,24,25,26]. The last two stored columns arepinky_CMCandthumb_MCP_AA. The previous thumb CMC_AA/MCP_FE/IP labels (26/2/3) are corrected to 2/3/4; the observation columns retain their original values. supplementary_joints:(T,5)in the orderLift, torso_flip, head_j1, head_j2, head_j3. Lift is in meters; the other four dimensions are in radians. Both Sharpa hands retain all 44 physical joint coordinates in their respective hand arrays. Seejoint_mapping.jsonandsupplementary_joints_note.json.SharpaTacMapSim: per hand(T,5,24,24)uint8 grayscale synthetic deformation images; original 0.015 m maximum-distance quantization is retained.IsaacPhysXContactForce: per hand(T,5,3)float32 world-frame force[Fx,Fy,Fz], in newtons.- Tactile areas per hand: thumb/index/middle/ring/little fingertips map to
0/1/2/3/4. No left-hand global offset is baked into the stored values. - Tactile
data/area/sensor/typefields are paired.supplementary_tactile_note.jsondocuments these standard fingertip groups; there is nosupplementary_tactilearray. - Camera and wrist poses are
(T,6)world-frame[x,y,z,rx,ry,rz]: translation in meters and rotation vector in radians, not Euler angles. - Original replay observations, controls, object poses, physical timestamps and pose quality markers are retained in
dextraverse_*arrays.
Timing and pose limitations
timestamps follows the exported video clock (20 fps, frame/20 seconds); dextraverse_sim_timestamps follows the physical replay clock (30 Hz, frame/30 seconds).
Do not use the video clock as the physical simulation sampling rate.
Camera and wrist poses were reconstructed from recorded replay joint states, the simulation URDF, base transforms and camera look-at parameters.
All 300 trajectories have an estimated first-frame camera pose because warm-up camera state was not recorded.
Thirty trajectories have a first-frame tactile pose reference discrepancy; 135,704 of 135,734 frames pass the independent pose reference check.
The data include explicit quality flags. Actual RGB exposure timestamps were not recorded, so exact exposure-level synchronization cannot be established from this export.
See documentation/pose_4d/README.md and its numerical validation report before use.
Mapping provenance and review status
The 22-axis mapping is an engineering correspondence derived from the official UniDex robot mapping examples and the Sharpa reconstruction URDF. Wuji's four serial thumb axes provide slots 1/2/3/4; Shadow's thumb middle abduction axis provides slot 26 and its fifth-finger metacarpal axis provides slot 25. These are existing semantic slots, not newly assigned reserved IDs. This is not an organizer-published Sharpa mapping or an OXT acceptance certificate. On 2026-10-05 the contributor accepted the remaining documented materials, including this mapping, without requesting changes. This is contributor acceptance, not an additional independent mechanical-axis validation.
See canonical_mapping_evidence.json for fixed source references, rationale and review status. Numerical verification proves lossless regrouping and agreement between stored indices and the supplied mapping; it does not prove anatomical equivalence solely from array shapes. Original model signs and zero positions are retained, with no undocumented sign correction or angle offset.
Authors, in contributor-specified order: xiangyu chen, chuhao zhou, jiaxiao shi.
Contact: xiangyu.chen@ntu.edu.sg; chuhao002@e.ntu.edu.sg; jiaxiao.shi@ntu.edu.sg.
The approved dataset version remains 1.0.0; this local revision is identified as sharpa22_supp5.
The contributor-provided NTU Singapore logo is included in the separate organizer submission package.
Paper, project website and code_trigger remain empty at the contributor's request.
An interactive GPT-6 Astra/xhigh visual review inspected 24 ego RGB/tactile montages (384 paired samples, 214 distinct task frames, 127 trajectories). The complete and compact JSON reports are included in the separate organizer package. Additional montage images, response exports, session provenance and contributor acceptance records are retained in the local audit materials; they are omitted from the minimal organizer package at the contributor's request. This was not an unmodified official CLI/API run. Original model warning scores and the timing, pose, rendering and coverage limitations remain unchanged; organizer acceptance of the disclosed methods and limitations is pending. Earlier mapping evidence files preserve their original engineering-review status; the contributor's later acceptance does not alter the underlying evidence.
This revision contains the updated 22/5 task archives and joint-index diagram. Use the archive SHA-256 values in SHA256SUMS.txt to distinguish these files from the earlier 20/9 layout; the dataset URL and version number remain unchanged.
- Downloads last month
- 708