The dataset viewer is not available for this split.
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
bytes: int64
id: string
path: string
sha256: string
control_hz: int64
status: string
test: int64
inventory_episodes: int64
validation: int64
configured_tasks: int64
tasks: int64
pending_replacement: int64
dataset: string
episodes: int64
train: int64
target_train_episodes: int64
observation_hz: struct<CP01: list<item: double>, CP02: list<item: double>, CP03: list<item: double>, CP04: list<item (... 275 chars omitted)
child 0, CP01: list<item: double>
child 0, item: double
child 1, CP02: list<item: double>
child 0, item: double
child 2, CP03: list<item: double>
child 0, item: double
child 3, CP04: list<item: double>
child 0, item: double
child 4, CP05: list<item: double>
child 0, item: double
child 5, E: list<item: int64>
child 0, item: int64
child 6, F: list<item: double>
child 0, item: double
child 7, FR1: list<item: double>
child 0, item: double
child 8, FR2: list<item: double>
child 0, item: double
child 9, M: list<item: double>
child 0, item: double
child 10, P: list<item: double>
child 0, item: double
child 11, PR1: list<item: double>
child 0, item: double
child 12, PR2: list<item: double>
child 0, item: double
child 13, PR3: list<item: double>
child 0, item: double
child 14, S: list<item: double>
child 0, item: double
counts: struct<CP01: struct<easy: int64, medium: int64, hard: int64>, CP02: struct<easy: int64, medium: int6 (... 711 chars omitted)
ch
...
m: int64
child 2, hard: int64
child 5, F: struct<easy: int64, medium: int64, hard: int64>
child 0, easy: int64
child 1, medium: int64
child 2, hard: int64
child 6, FR1: struct<easy: int64, medium: int64, hard: int64>
child 0, easy: int64
child 1, medium: int64
child 2, hard: int64
child 7, FR2: struct<easy: int64, medium: int64, hard: int64>
child 0, easy: int64
child 1, medium: int64
child 2, hard: int64
child 8, M: struct<easy: int64, medium: int64, hard: int64>
child 0, easy: int64
child 1, medium: int64
child 2, hard: int64
child 9, P: struct<easy: int64, medium: int64, hard: int64>
child 0, easy: int64
child 1, medium: int64
child 2, hard: int64
child 10, PR1: struct<easy: int64, medium: int64, hard: int64>
child 0, easy: int64
child 1, medium: int64
child 2, hard: int64
child 11, PR2: struct<easy: int64, medium: int64, hard: int64>
child 0, easy: int64
child 1, medium: int64
child 2, hard: int64
child 12, PR3: struct<easy: int64, medium: int64, hard: int64>
child 0, easy: int64
child 1, medium: int64
child 2, hard: int64
child 13, S: struct<easy: int64, medium: int64, hard: int64>
child 0, easy: int64
child 1, medium: int64
child 2, hard: int64
child 14, E: struct<easy: int64, medium: int64, hard: int64>
child 0, easy: int64
child 1, medium: int64
child 2, hard: int64
to
{'dataset': Value('string'), 'status': Value('string'), 'tasks': Value('int64'), 'episodes': Value('int64'), 'train': Value('int64'), 'validation': Value('int64'), 'test': Value('int64'), 'counts': {'CP01': {'easy': Value('int64'), 'medium': Value('int64'), 'hard': Value('int64')}, 'CP02': {'easy': Value('int64'), 'medium': Value('int64'), 'hard': Value('int64')}, 'CP03': {'easy': Value('int64'), 'medium': Value('int64'), 'hard': Value('int64')}, 'CP04': {'easy': Value('int64'), 'medium': Value('int64'), 'hard': Value('int64')}, 'CP05': {'easy': Value('int64'), 'medium': Value('int64'), 'hard': Value('int64')}, 'F': {'easy': Value('int64'), 'medium': Value('int64'), 'hard': Value('int64')}, 'FR1': {'easy': Value('int64'), 'medium': Value('int64'), 'hard': Value('int64')}, 'FR2': {'easy': Value('int64'), 'medium': Value('int64'), 'hard': Value('int64')}, 'M': {'easy': Value('int64'), 'medium': Value('int64'), 'hard': Value('int64')}, 'P': {'easy': Value('int64'), 'medium': Value('int64'), 'hard': Value('int64')}, 'PR1': {'easy': Value('int64'), 'medium': Value('int64'), 'hard': Value('int64')}, 'PR2': {'easy': Value('int64'), 'medium': Value('int64'), 'hard': Value('int64')}, 'PR3': {'easy': Value('int64'), 'medium': Value('int64'), 'hard': Value('int64')}, 'S': {'easy': Value('int64'), 'medium': Value('int64'), 'hard': Value('int64')}, 'E': {'easy': Value('int64'), 'medium': Value('int64'), 'hard': Value('int64')}}, 'control_hz': Value('int64'), 'observation_hz': {'CP01': List(Value('float64')), 'CP02': List(Value('float64')), 'CP03': List(Value('float64')), 'CP04': List(Value('float64')), 'CP05': List(Value('float64')), 'E': List(Value('int64')), 'F': List(Value('float64')), 'FR1': List(Value('float64')), 'FR2': List(Value('float64')), 'M': List(Value('float64')), 'P': List(Value('float64')), 'PR1': List(Value('float64')), 'PR2': List(Value('float64')), 'PR3': List(Value('float64')), 'S': List(Value('float64'))}, 'configured_tasks': Value('int64'), 'inventory_episodes': Value('int64'), 'target_train_episodes': Value('int64'), 'pending_replacement': Value('int64')}
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
bytes: int64
id: string
path: string
sha256: string
control_hz: int64
status: string
test: int64
inventory_episodes: int64
validation: int64
configured_tasks: int64
tasks: int64
pending_replacement: int64
dataset: string
episodes: int64
train: int64
target_train_episodes: int64
observation_hz: struct<CP01: list<item: double>, CP02: list<item: double>, CP03: list<item: double>, CP04: list<item (... 275 chars omitted)
child 0, CP01: list<item: double>
child 0, item: double
child 1, CP02: list<item: double>
child 0, item: double
child 2, CP03: list<item: double>
child 0, item: double
child 3, CP04: list<item: double>
child 0, item: double
child 4, CP05: list<item: double>
child 0, item: double
child 5, E: list<item: int64>
child 0, item: int64
child 6, F: list<item: double>
child 0, item: double
child 7, FR1: list<item: double>
child 0, item: double
child 8, FR2: list<item: double>
child 0, item: double
child 9, M: list<item: double>
child 0, item: double
child 10, P: list<item: double>
child 0, item: double
child 11, PR1: list<item: double>
child 0, item: double
child 12, PR2: list<item: double>
child 0, item: double
child 13, PR3: list<item: double>
child 0, item: double
child 14, S: list<item: double>
child 0, item: double
counts: struct<CP01: struct<easy: int64, medium: int64, hard: int64>, CP02: struct<easy: int64, medium: int6 (... 711 chars omitted)
ch
...
m: int64
child 2, hard: int64
child 5, F: struct<easy: int64, medium: int64, hard: int64>
child 0, easy: int64
child 1, medium: int64
child 2, hard: int64
child 6, FR1: struct<easy: int64, medium: int64, hard: int64>
child 0, easy: int64
child 1, medium: int64
child 2, hard: int64
child 7, FR2: struct<easy: int64, medium: int64, hard: int64>
child 0, easy: int64
child 1, medium: int64
child 2, hard: int64
child 8, M: struct<easy: int64, medium: int64, hard: int64>
child 0, easy: int64
child 1, medium: int64
child 2, hard: int64
child 9, P: struct<easy: int64, medium: int64, hard: int64>
child 0, easy: int64
child 1, medium: int64
child 2, hard: int64
child 10, PR1: struct<easy: int64, medium: int64, hard: int64>
child 0, easy: int64
child 1, medium: int64
child 2, hard: int64
child 11, PR2: struct<easy: int64, medium: int64, hard: int64>
child 0, easy: int64
child 1, medium: int64
child 2, hard: int64
child 12, PR3: struct<easy: int64, medium: int64, hard: int64>
child 0, easy: int64
child 1, medium: int64
child 2, hard: int64
child 13, S: struct<easy: int64, medium: int64, hard: int64>
child 0, easy: int64
child 1, medium: int64
child 2, hard: int64
child 14, E: struct<easy: int64, medium: int64, hard: int64>
child 0, easy: int64
child 1, medium: int64
child 2, hard: int64
to
{'dataset': Value('string'), 'status': Value('string'), 'tasks': Value('int64'), 'episodes': Value('int64'), 'train': Value('int64'), 'validation': Value('int64'), 'test': Value('int64'), 'counts': {'CP01': {'easy': Value('int64'), 'medium': Value('int64'), 'hard': Value('int64')}, 'CP02': {'easy': Value('int64'), 'medium': Value('int64'), 'hard': Value('int64')}, 'CP03': {'easy': Value('int64'), 'medium': Value('int64'), 'hard': Value('int64')}, 'CP04': {'easy': Value('int64'), 'medium': Value('int64'), 'hard': Value('int64')}, 'CP05': {'easy': Value('int64'), 'medium': Value('int64'), 'hard': Value('int64')}, 'F': {'easy': Value('int64'), 'medium': Value('int64'), 'hard': Value('int64')}, 'FR1': {'easy': Value('int64'), 'medium': Value('int64'), 'hard': Value('int64')}, 'FR2': {'easy': Value('int64'), 'medium': Value('int64'), 'hard': Value('int64')}, 'M': {'easy': Value('int64'), 'medium': Value('int64'), 'hard': Value('int64')}, 'P': {'easy': Value('int64'), 'medium': Value('int64'), 'hard': Value('int64')}, 'PR1': {'easy': Value('int64'), 'medium': Value('int64'), 'hard': Value('int64')}, 'PR2': {'easy': Value('int64'), 'medium': Value('int64'), 'hard': Value('int64')}, 'PR3': {'easy': Value('int64'), 'medium': Value('int64'), 'hard': Value('int64')}, 'S': {'easy': Value('int64'), 'medium': Value('int64'), 'hard': Value('int64')}, 'E': {'easy': Value('int64'), 'medium': Value('int64'), 'hard': Value('int64')}}, 'control_hz': Value('int64'), 'observation_hz': {'CP01': List(Value('float64')), 'CP02': List(Value('float64')), 'CP03': List(Value('float64')), 'CP04': List(Value('float64')), 'CP05': List(Value('float64')), 'E': List(Value('int64')), 'F': List(Value('float64')), 'FR1': List(Value('float64')), 'FR2': List(Value('float64')), 'M': List(Value('float64')), 'P': List(Value('float64')), 'PR1': List(Value('float64')), 'PR2': List(Value('float64')), 'PR3': List(Value('float64')), 'S': List(Value('float64'))}, 'configured_tasks': Value('int64'), 'inventory_episodes': Value('int64'), 'target_train_episodes': Value('int64'), 'pending_replacement': Value('int64')}
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.
Mind2Act shared assets
This repository uses a shared Assets/ root, following RoboDojo's asset categories. Robots are shared by robot type, never copied into each task directory. Source files, task rules, controller implementations and scorer bindings live in the current merged code branch. Training trajectories, evaluation answer sequences and checkpoints remain separate releases.
Assets/
Robots/x5/ # ARX USD, URDF, meshes and planning YAML
profiles/ # explicitly named task physical overrides
Object/MemBench/{Articulation,Rigid,Geometry}/<object>/00000/
object.usd + metadata.json
Material/{MemBench,NVIDIA}/
Background/MemBench/cp05_station/00000/station.usdc
Sensor/MemBench/camera_stand/00000/object.usd
Eval_Layout/MemBench/arx_x5/CP05/{easy,medium,hard}/layout.json
manifests/
robots/x5.json
tasks/CP05.json
previews/CP05/cp05-assets-20261004-v2/
asset-manifest.json
object.usd may contain binary USDC data; the extension matches the official loader's USD fallback. A single piano asset contains distinct keys and joints, not one fused mesh. 00000 identifies a model variant, not a seed or release. Releases are selected by immutable HF commit and component-manifest SHA256 in GitHub's configs/assets/lock.json.
The X5 base asset is unchanged. CP05's pre-existing robot physical overrides were extracted into the explicit cp05-contact-repair-v18 profile layer; station mounts, HOME/state and appearance remain in the station. This is not adoption of CP01's controller profile. The shared maintenance implementation is registered in GitHub's configs/physco/x5_runtime_version.json; CP01's limited evidence does not qualify other tasks. No gentle profile is automatically replaced.
This aligns directory categories and object entrypoints with RoboDojo. Mind2Act component manifests, extensions to object metadata, and the CP05 layout/task adapter are explicitly additional interfaces; no drop-in whole-task RoboDojo equivalence is claimed. Functional points are not invented: metadata leaves unqualified active/passive functions empty.
The selected CP05 release is cp05-assets-20261004-v2. Its qualification scope is recorded in manifests/tasks/CP05.json. Task demonstration/green feedback/scoring require code; asset reset qualification alone is not full task acceptance. The prior package remains recoverable at commit da7f926ee8762a68570c97e5ddce7c00a03ca6ea and tag cp05-task-package-v1; old packages/ paths are retired from current main.
Source-specific third-party attribution and public redistribution review are recorded in the code repository. Both repositories remain private; this README grants no blanket third-party license.
CP01 selected final assets
CP01 reuses the unchanged Assets/Robots/x5 component. Its current station, dependency assets and 150 per-seed JSON layouts (50 per difficulty) are indexed by manifests/tasks/CP01.json. The manifest and all asset bytes are identical to the selected local/native final library; the recorded shared HF revision inside this manifest is provenance for the shared X5 baseline, not the revision of this new CP01 publication. GitHub pins the full publication commit separately.
The 150 layouts passed CPU checks; one seed per difficulty passed native reset/reference/hold, totaling 7008 ticks. Full-task success and all-150 native qualification are not established (formal_eligible=false). G05 uses joint actions; the separate Astra/move_eef planner is not qualified by this publication. Private CP01 layouts include environment parameters and binding metadata; they are not an answer-free public evaluation release. Historical source and results remain unchanged.
CP03 / CP04 selected final assets
manifests/tasks/CP03.json and manifests/tasks/CP04.json index the selected cp03-final and cp04-final components. Both reuse the unchanged shared X5 component. GitHub pins this publication commit; original shared-base revisions inside manifests record provenance, not the new task publication revision. Asset manifest and payload bytes match their selected local libraries. Historical and rejected candidates are not selectable entries.
Each task has 150 frozen TEST bindings (50 per difficulty). CP03 passed three representative native reset/render/hold and both-arm small joint playback checks, 880 ticks each; its videos include a labelled one-second final hold. CP03's three collector-report files were sanitized after those native probes: speed is unchanged, original hashes are retained, and all other asset bytes are unchanged. Both pre- and post-sanitization manifest identities remain in its qualification records.
CP04 passed three representative native reset/render/robot readback/240-tick hold checks, plus 12 initialization ticks each. This does not qualify grasping, drawer interaction, cap removal, pouring, heating or its complete task controller. Its selected task-local arm/finger parameters are preserved; shared command/observation adoption remains pending.
These are asset publications, not full-task or all-150 native qualification (formal_eligible=false). No model requests or new native simulation were issued for publication. Build-time manifest qualification flags remain immutable; later, limited qualification is in the task selection and adoption receipts. This repository stays private.
| Task | Easy | Medium | Hard |
|---|---|---|---|
| CP03 |  |
 |
 |
| CP04 |  |
 |
 |
CP02 selected current assets
manifests/tasks/CP02.json preserves the original selected 43-file closure. Cake, macarons, box/lid, buttons and Hard cover meshes are embedded in the three templates/CP02/<difficulty>/scene.usda layers; shared X5, sensor and material paths are reused without changing bytes. GitHub contains the loader, explicit cp02-gentle profile and 150 fixed TEST hash bindings. These are assets, not training trajectories.
The sealed C source passed three-level reset/readback, both-arm free-space/finger checks and Medium/Hard physical lid/Start checks. Full macaron grasp/placement, Hard cover placement and complete collision-world qualification remain pending. An audit found that its template-only materializer omitted per-seed target/source permutations and initial cake phase. The publication code repairs those bindings and preserves the original receipts; the repaired bindings and ported controller still require new native qualification. No model requests or new native simulations were launched (formal_eligible=false).
The following images are final frames of the historical zero-model native controls probes, not new fixed-seed initial frames or model rollouts:
| Easy | Medium | Hard |
|---|---|---|
 |
 |
 |
Artifact storage index
management/storage-index.json separates published assets, the incomplete legacy training upload, and the current 2,250-episode server training selection. The training metadata stored under management/training/ is an inventory only: no training HDF5 payloads are included in this asset repository. Current runtime selections and immutable component pins remain unchanged.
Current merged case resources
The E/P/S/F/M, FR1/FR2 and PR1/PR2/PR3 resource manifests are in manifests/tasks/. Shared model and material resources are indexed by manifests/resources/common.json. Files are copied byte-for-byte from the current merged asset workspace; seed-specific scenes are authored by the task code. These resources do not include historical code snapshots, review videos, debug logs, training data or checkpoints. Runtime MDL dependencies and existing qualification limits remain explicit in each manifest. CP01–CP05 retain their existing asset components; CP02 replacement is reviewed separately.
CP02 current geometry
Easy, Medium and Hard share the approved 35 mm rear box extension. Side walls are 190 mm deep and the lid panel is 183 mm deep. The extension is baked into the existing template files; the current CP02 loader validates it and does not extend it again. The three action-start renders were reviewed and approved. This publication does not claim complete task, grasp or collision qualification.
Access and matching code
This asset repository is public; downloading does not require a personal token. The current merged code is on checkpoint/20261014-pre-eval in walawalagoose/Mind2Act-Eval; the code repository remains private and requires its own access. CP02 selects the approved asset commit b216daf877a3cdc8193a9fdae5727e05d6705530. Install the documented Isaac Sim environment and dependencies, and use the code-selected asset commit and manifests. Publication does not certify every case as a standalone portable installation: existing unresolved resource references remain listed in the non-CP resource manifests.
- Downloads last month
- 983