The dataset viewer is not available for this split.
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
environment_count: int64
episodes: int64
file_count: int64
format_version: int64
generation_runtime_record: struct<machine: string, packages: struct<crafter: string, huggingface_hub: string, imageio: string, (... 312 chars omitted)
child 0, machine: string
child 1, packages: struct<crafter: string, huggingface_hub: string, imageio: string, numpy: string, opensimplex: string (... 54 chars omitted)
child 0, crafter: string
child 1, huggingface_hub: string
child 2, imageio: string
child 3, numpy: string
child 4, opensimplex: string
child 5, pillow: string
child 6, pyyaml: string
child 7, ruamel.yaml: string
child 2, platform: string
child 3, python: string
child 4, scope: string
child 5, thread_limits: struct<MKL_NUM_THREADS: string, NUMEXPR_NUM_THREADS: string, OMP_NUM_THREADS: string, OPENBLAS_NUM_T (... 15 chars omitted)
child 0, MKL_NUM_THREADS: string
child 1, NUMEXPR_NUM_THREADS: string
child 2, OMP_NUM_THREADS: string
child 3, OPENBLAS_NUM_THREADS: string
child 6, zlib_build: string
child 7, zlib_runtime: string
payload_bytes: int64
release_verification_runtime: struct<packages: struct<crafter: string, huggingface_hub: string, numpy: string, procgen: null, pyya (... 70 chars omitted)
child 0, packages: struct<crafter: string, huggingface_hub: string, numpy: string, procgen: null, pyyaml: string>
child 0, crafter: string
child 1, huggingface_hub: string
child 2, num
...
ta_colu (... 29 chars omitted)
child 0, item: struct<label: string, label_status: string, local_id: int64, source_id: int64, delta_column: int64, (... 17 chars omitted)
child 0, label: string
child 1, label_status: string
child 2, local_id: int64
child 3, source_id: int64
child 4, delta_column: int64
child 5, delta_row: int64
child 2, env_name: string
child 3, label_status: string
child 4, manifest: string
child 5, manifest_sha256: string
child 6, num_actions: int64
child 7, source: struct<kind: string, package: string, symbol: string, version: string, data_factory_file: string, fi (... 11 chars omitted)
child 0, kind: string
child 1, package: string
child 2, symbol: string
child 3, version: string
child 4, data_factory_file: string
child 5, file: string
child 8, source_action_count: int64
child 9, subset: string
child 10, coordinates: struct<column: string, delta_order: list<item: string>, delta_units: string, movement_rule: string, (... 40 chars omitted)
child 0, column: string
child 1, delta_order: list<item: string>
child 0, item: string
child 2, delta_units: string
child 3, movement_rule: string
child 4, observation_layout: string
child 5, row: string
scope: string
action_id_semantics: string
to
{'action_id_semantics': Value('string'), 'format_version': Value('int64'), 'scope': Value('string'), 'spaces': List({'action_subset': Value('null'), 'actions': List({'label': Value('string'), 'label_status': Value('string'), 'local_id': Value('int64'), 'source_id': Value('int64'), 'delta_column': Value('int64'), 'delta_row': Value('int64')}), 'env_name': Value('string'), 'label_status': Value('string'), 'manifest': Value('string'), 'manifest_sha256': Value('string'), 'num_actions': Value('int64'), 'source': {'kind': Value('string'), 'package': Value('string'), 'symbol': Value('string'), 'version': Value('string'), 'data_factory_file': Value('string'), 'file': Value('string')}, 'source_action_count': Value('int64'), 'subset': Value('string'), 'coordinates': {'column': Value('string'), 'delta_order': List(Value('string')), 'delta_units': Value('string'), 'movement_rule': Value('string'), 'observation_layout': Value('string'), 'row': Value('string')}})}
because column names don't match
Traceback: Traceback (most recent call last):
File "/src/services/worker/src/worker/utils.py", line 149, 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 129, 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 544, 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 3052, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2559, in __iter__
for key, pa_table in self._iter_arrow():
~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2584, 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 558, in _iter_arrow
for key, pa_table in iterator:
^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 440, 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 361, 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 2381, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2309, in cast_table_to_schema
raise CastError(
...<3 lines>...
)
datasets.table.CastError: Couldn't cast
environment_count: int64
episodes: int64
file_count: int64
format_version: int64
generation_runtime_record: struct<machine: string, packages: struct<crafter: string, huggingface_hub: string, imageio: string, (... 312 chars omitted)
child 0, machine: string
child 1, packages: struct<crafter: string, huggingface_hub: string, imageio: string, numpy: string, opensimplex: string (... 54 chars omitted)
child 0, crafter: string
child 1, huggingface_hub: string
child 2, imageio: string
child 3, numpy: string
child 4, opensimplex: string
child 5, pillow: string
child 6, pyyaml: string
child 7, ruamel.yaml: string
child 2, platform: string
child 3, python: string
child 4, scope: string
child 5, thread_limits: struct<MKL_NUM_THREADS: string, NUMEXPR_NUM_THREADS: string, OMP_NUM_THREADS: string, OPENBLAS_NUM_T (... 15 chars omitted)
child 0, MKL_NUM_THREADS: string
child 1, NUMEXPR_NUM_THREADS: string
child 2, OMP_NUM_THREADS: string
child 3, OPENBLAS_NUM_THREADS: string
child 6, zlib_build: string
child 7, zlib_runtime: string
payload_bytes: int64
release_verification_runtime: struct<packages: struct<crafter: string, huggingface_hub: string, numpy: string, procgen: null, pyya (... 70 chars omitted)
child 0, packages: struct<crafter: string, huggingface_hub: string, numpy: string, procgen: null, pyyaml: string>
child 0, crafter: string
child 1, huggingface_hub: string
child 2, num
...
ta_colu (... 29 chars omitted)
child 0, item: struct<label: string, label_status: string, local_id: int64, source_id: int64, delta_column: int64, (... 17 chars omitted)
child 0, label: string
child 1, label_status: string
child 2, local_id: int64
child 3, source_id: int64
child 4, delta_column: int64
child 5, delta_row: int64
child 2, env_name: string
child 3, label_status: string
child 4, manifest: string
child 5, manifest_sha256: string
child 6, num_actions: int64
child 7, source: struct<kind: string, package: string, symbol: string, version: string, data_factory_file: string, fi (... 11 chars omitted)
child 0, kind: string
child 1, package: string
child 2, symbol: string
child 3, version: string
child 4, data_factory_file: string
child 5, file: string
child 8, source_action_count: int64
child 9, subset: string
child 10, coordinates: struct<column: string, delta_order: list<item: string>, delta_units: string, movement_rule: string, (... 40 chars omitted)
child 0, column: string
child 1, delta_order: list<item: string>
child 0, item: string
child 2, delta_units: string
child 3, movement_rule: string
child 4, observation_layout: string
child 5, row: string
scope: string
action_id_semantics: string
to
{'action_id_semantics': Value('string'), 'format_version': Value('int64'), 'scope': Value('string'), 'spaces': List({'action_subset': Value('null'), 'actions': List({'label': Value('string'), 'label_status': Value('string'), 'local_id': Value('int64'), 'source_id': Value('int64'), 'delta_column': Value('int64'), 'delta_row': Value('int64')}), 'env_name': Value('string'), 'label_status': Value('string'), 'manifest': Value('string'), 'manifest_sha256': Value('string'), 'num_actions': Value('int64'), 'source': {'kind': Value('string'), 'package': Value('string'), 'symbol': Value('string'), 'version': Value('string'), 'data_factory_file': Value('string'), 'file': Value('string')}, 'source_action_count': Value('int64'), 'subset': Value('string'), 'coordinates': {'column': Value('string'), 'delta_order': List(Value('string')), 'delta_units': Value('string'), 'movement_rule': Value('string'), 'observation_layout': Value('string'), 'row': Value('string')}})}
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.
WM Interaction v1
Seeded synthetic interaction trajectories containing observation, action and next observation.
12000 episodes; 818161 transitions; 3 environments across 3 subsets.
Usage from this dataset directory: import numpy as np; episode = np.load('crafter/episode_000000.npz', allow_pickle=False)
Download the complete snapshot
The complete release is distributed as wm-interaction-v1-dataset.zip, containing all 12016 delivery files. Extract it and enter the wm-interaction-v1 directory before using the NPZ example above.
Archive: 822009439 bytes; SHA256: 63d7b99196b34f8c3611a4a34fb57d159aa55cbcbb2022f4b9c0858878e095d0. The per-file SHA256SUMS and all three subsets are inside the archive. The inventory below describes the unpacked delivery, rather than repository transport files.
from huggingface_hub import hf_hub_download
from zipfile import ZipFile
path = hf_hub_download('KineJIng/wm-interaction-v1', 'wm-interaction-v1-dataset.zip', repo_type='dataset')
with ZipFile(path) as archive:
archive.extractall('data')
Actual inventory
The delivery contains 12016 files and 819828703 bytes, including this card and SHA256SUMS. The data payload, excluding generated release documents and checksums, is 818699069 bytes.
crafter
Environment: crafter; difficulty: None; episodes: 2000; transitions: 338161; mean episode length: 169.0805.
Implementation variant: Crafter canonical-object-order variant.
Observation shape (HWC): [64, 64, 3]; categorical actions: 17; action subset mapping: None.
NPZ bytes: 781186604; manifest bytes: 693658; manifest SHA256: 62130b9baf7c4d0ead2b1ba2154da5e42d5ea9c613cf9169132efd1d6335602a.
Generation command (PowerShell, run from the wmbench project):
py -3.12 -m wmbench.data.collect --env 'crafter' --episodes 2000 --seed 0 --out 'data/v1/crafter' --env-kwargs '{"canonical_object_order":true}'
Crafter has no native easy/medium/hard difficulty setting. Its native episode limit is 10,000 steps unless an explicit environment option records otherwise. The canonical-object-order variant sorts candidate objects before upstream object balancing to remove object-set iteration-order differences between processes; it must not be represented as an unmodified upstream Crafter rollout.
Action sensitivity was evaluated at 16000/338161 recorded steps (coverage 0.0473147406117; sampled: true).
Sampling method: evenly spaced integer indices including endpoints; one sample uses step 0. Low-score threshold: 1; low-score fraction: 0.391875.
Variance/mean distribution: {"mean": 6.679823896624033, "median": 1.8630011144415697, "p10": 0.0, "p90": 24.44402261778412}.
Pairwise next-observation L1 distribution: {"mean": 2.9561477131563074, "median": 2.0344734939874387, "p10": 0.0, "p90": 7.277147001378678}.
Metric report bound to the manifest SHA256: true.
toy
Environment: toy; difficulty: easy; episodes: 5000; transitions: 160000; mean episode length: 32.
Implementation variant: wmbench data-only obstacle-navigation variant.
Observation shape (HWC): [8, 8, 3]; categorical actions: 4; action subset mapping: None.
NPZ bytes: 7688392; manifest bytes: 1562780; manifest SHA256: 847c94a3914391c25a8bd052ce4d0a620f9a66adedb5442228543a36c708d0b0.
Generation command (PowerShell, run from the wmbench project):
py -3.12 -m wmbench.data.collect --env 'toy' --episodes 5000 --seed 0 --out 'data/v1/toy' --difficulty 'easy'
Action sensitivity was evaluated at 160000/160000 recorded steps (coverage 1; sampled: false).
Sampling method: all steps. Low-score threshold: 1; low-score fraction: 0.8643875.
Variance/mean distribution: {"mean": 0.4656303873697917, "median": 0.6640625, "p10": 0.0, "p90": 1.328125}.
Pairwise next-observation L1 distribution: {"mean": 2.63610400390625, "median": 2.65625, "p10": 2.2135416666666665, "p90": 2.65625}.
Metric report bound to the manifest SHA256: true.
toy_hard
Environment: toy_hard; difficulty: hard; episodes: 5000; transitions: 320000; mean episode length: 64.
Implementation variant: wmbench data-only obstacle-navigation variant.
Observation shape (HWC): [16, 16, 3]; categorical actions: 4; action subset mapping: None.
NPZ bytes: 18461398; manifest bytes: 1597780; manifest SHA256: 72625a8df532dc3ae4bc79eaf77efb1537e1b7e0dcd384f576624356d5860919.
Generation command (PowerShell, run from the wmbench project):
py -3.12 -m wmbench.data.collect --env 'toy_hard' --episodes 5000 --seed 0 --out 'data/v1/toy_hard' --difficulty 'hard'
Action sensitivity was evaluated at 320000/320000 recorded steps (coverage 1; sampled: false).
Sampling method: all steps. Low-score threshold: 1; low-score fraction: 1.
Variance/mean distribution: {"mean": 0.19279854532877605, "median": 0.166015625, "p10": 0.0, "p90": 0.33203125}.
Pairwise next-observation L1 distribution: {"mean": 0.6142775268554688, "median": 0.6640625, "p10": 0.5533854166666666, "p90": 0.6640625}.
Metric report bound to the manifest SHA256: true.
Data schema
Each episode NPZ contains obs and next_obs as (T,H,W,C) uint8 arrays, action as (T,) little-endian int64, reward as (T,) little-endian float64, and separate boolean done and truncated arrays. Only the final transition is terminal or truncated. Consecutive observations and all manifest file hashes are verified.
meta.jsonl records episode ID, environment, actual reset seed, length, image shape, action count and NPZ SHA256, plus collection seed and configured generation options. When supplied, action_spaces.json maps local action IDs to source IDs and source-defined labels. The collector samples allowed actions uniformly at random; these are not expert demonstrations.
Metric definition and limits
At each analyzed time index, the complete replayed environment state, including its local RNG, is copied before branching all actions. For action a, magnitude m(a) is the mean absolute uint8 pixel-channel difference from the current observation. The requested score is population variance(m) / mean(m), or zero when mean(m) is zero. Equal-magnitude changes in different directions can produce a zero score despite distinct action effects. The pairwise next-observation L1 diagnostic is reported separately. A low score alone does not establish that data is causally useless. The raw pixel-mean score and its illustrative threshold depend on observation scale and resolution; do not use one threshold as a cross-environment rejection rule. Sampled metric statistics apply to the disclosed analyzed time indices, not an unmeasured full population.
Reproducibility
Environment reset seed equals collection seed plus episode ID. Action sampling uses PCG64 with SeedSequence [collection_seed, episode_id, 0x574D4244]. Episode archives fix member order, ZIP timestamps and permissions, NPY version, numeric byte order and DEFLATE level. SHA256SUMS covers every delivered file except itself.
Byte-identical regeneration requires the same collector and environment implementations, parameters, dependencies and runtime. The runtime below was inspected during release verification; it is not independent evidence of the earlier generation runtime. A separately supplied runtime.json is included as a generation record when present. A matching hash validates bytes, not downstream model performance.
When supplied, runtime.json and source_sha256.json record the actual generation dependency versions and source fingerprints. They are included in SHA256SUMS and must accompany regeneration.
{
"packages": {
"crafter": "1.8.3",
"huggingface_hub": "2.3.0",
"numpy": "2.5.3",
"procgen": null,
"pyyaml": "6.0.3"
},
"python": "3.12.10",
"zlib_build": "1.3.1",
"zlib_runtime": "1.3.1"
}
Known limitations
These are synthetic 2D game/navigation interactions. They contain no real robot recordings, camera calibration, continuous actuator commands, force/torque signals or demonstrated sim-to-real transfer. Random-action collection does not guarantee task success, balanced event coverage, semantic diversity or training gains. No world-model or robot policy performance claim is made. Train/evaluation splits should use disjoint seeds and be assessed for state overlap before a training experiment.
Sources and distribution terms
Toy subsets are generated from the local wmbench toy/data-only obstacle-navigation implementation. The release records its actual environment options; no external video or robot trajectory is incorporated into those subsets. KineJing's original trajectory records, action annotations, metadata and compilation contributions in this public reference release are licensed under Creative Commons Attribution 4.0 International (CC BY 4.0), within the scope stated in LICENSE-DATA.md. Commercial reuse is permitted within that license's scope. No new source-code license is granted by this dataset card. Paid services cover customer-environment collection, counterfactual quality checks, reproducibility acceptance and delivery.
Crafter, when included, is derived from Danijar Hafner's MIT-licensed project: https://github.com/danijar/crafter . The upstream license is at https://github.com/danijar/crafter/blob/main/LICENSE . Its package includes graphical assets; no exclusive ownership of third-party code, assets or rendered pixels is claimed. Keep the upstream copyright and MIT permission notice in THIRD_PARTY_NOTICES.txt. The license: other metadata denotes CC BY 4.0 for original contributions together with the retained upstream MIT terms; it does not replace or broaden upstream rights. CC BY 4.0 terms: https://creativecommons.org/licenses/by/4.0/ .
- Downloads last month
- 72