Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
calls_completed: bool
calls: list<item: struct<index: int64, skill: string, arguments: struct<target: string, destination: string (... 393 chars omitted)
  child 0, item: struct<index: int64, skill: string, arguments: struct<target: string, destination: string, receptacl (... 381 chars omitted)
      child 0, index: int64
      child 1, skill: string
      child 2, arguments: struct<target: string, destination: string, receptacle: string, region: string>
          child 0, target: string
          child 1, destination: string
          child 2, receptacle: string
          child 3, region: string
      child 3, success: bool
      child 4, status: string
      child 5, start_time: double
      child 6, preconditions: struct<known_target: bool, hand_state: bool, empty_hand: bool, correct_fixture: bool, object_held: b (... 4 chars omitted)
          child 0, known_target: bool
          child 1, hand_state: bool
          child 2, empty_hand: bool
          child 3, correct_fixture: bool
          child 4, object_held: bool
      child 7, postconditions: struct<verified: bool>
          child 0, verified: bool
      child 8, end_time: double
      child 9, wall_seconds: double
      child 10, motion_timing: struct<simulated_seconds: double, target_seconds: list<item: double>, status: string>
          child 0, simulated_seconds: double
          child 1, target_seconds: list<item: double>
              child 0, item: double
          child 2, status: string
success: bool
goal_evidence: struct<bowls_in_cabinet: bool, cups_in_drawer: bool, cabinet_closed: bool, drawer_closed: bool, empt (... 13 chars omitted)
  child 0, bowls_in_cabinet: bool
  child 1, cups_in_drawer: bool
  child 2, cabinet_closed: bool
  child 3, drawer_closed: bool
  child 4, empty_hand: bool
setup_evidence: struct<objects_on_requested_supports: bool, empty_hand: bool, initial_robot_clear: bool, parameters_ (... 14 chars omitted)
  child 0, objects_on_requested_supports: bool
  child 1, empty_hand: bool
  child 2, initial_robot_clear: bool
  child 3, parameters_applied: bool
video: string
video_speedup: double
scene_spec_sha256: string
unavailable: list<item: null>
  child 0, item: null
selected: struct<bowl_01: string, bowl_02: string, cup_01: string, cup_02: string>
  child 0, bowl_01: string
  child 1, bowl_02: string
  child 2, cup_01: string
  child 3, cup_02: string
to
{'unavailable': List(Value('null')), 'selected': {'bowl_01': Value('string'), 'bowl_02': Value('string'), 'cup_01': Value('string'), 'cup_02': 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
              calls_completed: bool
              calls: list<item: struct<index: int64, skill: string, arguments: struct<target: string, destination: string (... 393 chars omitted)
                child 0, item: struct<index: int64, skill: string, arguments: struct<target: string, destination: string, receptacl (... 381 chars omitted)
                    child 0, index: int64
                    child 1, skill: string
                    child 2, arguments: struct<target: string, destination: string, receptacle: string, region: string>
                        child 0, target: string
                        child 1, destination: string
                        child 2, receptacle: string
                        child 3, region: string
                    child 3, success: bool
                    child 4, status: string
                    child 5, start_time: double
                    child 6, preconditions: struct<known_target: bool, hand_state: bool, empty_hand: bool, correct_fixture: bool, object_held: b (... 4 chars omitted)
                        child 0, known_target: bool
                        child 1, hand_state: bool
                        child 2, empty_hand: bool
                        child 3, correct_fixture: bool
                        child 4, object_held: bool
                    child 7, postconditions: struct<verified: bool>
                        child 0, verified: bool
                    child 8, end_time: double
                    child 9, wall_seconds: double
                    child 10, motion_timing: struct<simulated_seconds: double, target_seconds: list<item: double>, status: string>
                        child 0, simulated_seconds: double
                        child 1, target_seconds: list<item: double>
                            child 0, item: double
                        child 2, status: string
              success: bool
              goal_evidence: struct<bowls_in_cabinet: bool, cups_in_drawer: bool, cabinet_closed: bool, drawer_closed: bool, empt (... 13 chars omitted)
                child 0, bowls_in_cabinet: bool
                child 1, cups_in_drawer: bool
                child 2, cabinet_closed: bool
                child 3, drawer_closed: bool
                child 4, empty_hand: bool
              setup_evidence: struct<objects_on_requested_supports: bool, empty_hand: bool, initial_robot_clear: bool, parameters_ (... 14 chars omitted)
                child 0, objects_on_requested_supports: bool
                child 1, empty_hand: bool
                child 2, initial_robot_clear: bool
                child 3, parameters_applied: bool
              video: string
              video_speedup: double
              scene_spec_sha256: string
              unavailable: list<item: null>
                child 0, item: null
              selected: struct<bowl_01: string, bowl_02: string, cup_01: string, cup_02: string>
                child 0, bowl_01: string
                child 1, bowl_02: string
                child 2, cup_01: string
                child 3, cup_02: string
              to
              {'unavailable': List(Value('null')), 'selected': {'bowl_01': Value('string'), 'bowl_02': Value('string'), 'cup_01': Value('string'), 'cup_02': Value('string')}}
              because column names don't match

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.

Cross-Sim Data: Fine-Tuning and Evaluation

Successful simulated household composite demonstrations for task-specific fine-tuning and evaluation. This snapshot contains 147 episodes across five tasks: set table, clear and store, prepare and serve meal, unpack and sort groceries, and warm two pots.

Fine-tuning and evaluation stages

The fine-tuning stage adapts a pretrained VLA using the fine-tuning set. The evaluation stage measures the resulting policy on the two evaluation sets. The machine-readable train key and existing paths remain unchanged for archive/tool compatibility; they denote the fine-tuning set, not pretraining.

Fine-tuning set (train): two-room scenes with limited variations; target at least 60 successful minutes per task, with required procedure and door-opening coverage.
Two-room evaluation set (test_two_rooms): two-room scenes with the configured evaluation variations; target at least 15 successful minutes per task.
Multi-room evaluation set (test_multi_room): multiple-room scenes with the configured evaluation variations; target at least 15 successful minutes per task.

These are minimum duration targets, not exact caps. Whole episodes and concurrent workers produce overshoot. Durations are finalized 1× videos, including explicit task waits. Inspect episode configurations and asset_overlap.json for actual variation and asset overlap; this card does not assert every variation is disjoint.

Task	Fine-tuning minutes	Two-room evaluation minutes	Multi-room evaluation minutes
set_table	90.96	24.65	29.66
clear_and_store	74.33	70.60	36.84
prepare_and_serve_meal	84.82	29.02	28.07
unpack_and_sort_groceries	71.88	17.48	17.07
warm_two_pots	74.25	25.90	22.86

Files and use

episodes.jsonl indexes each episode archive, SHA-256, task, split, duration, and browser-accessible preview video and layout. previews/ also exposes prompts, configuration, skill calls, success reports, and asset selections.

Each archives/<split>/<task>/<episode>.tar.gz contains the complete accepted episode folder: video, scene/layout metadata, prompt, ordered skill calls, simulation trace, action recording, initial state and compiled models, along with the saved diagnostic metadata. Rejected speculative-trial directories and failed episodes are excluded. Source episodes are not modified or re-rendered.

Download with huggingface_hub.snapshot_download("plnguyen2908/cross-sim-data", repo_type="dataset"). Extract chosen archives into the downloaded dataset root, for example:

tar -xzf archives/train/set_table/EPISODE.tar.gz -C DATASET_ROOT

Archive members already include their original split/batch/episode paths. After extraction, relative paths in split manifests and per-task manifests under tasks/<task>/<split>/ resolve within the dataset. Layout indexes and object inventories are included. Preview copies duplicate a small subset for browsing.

The raw archive action recordings are simulator commands. Standard LeRobot v3 fine-tuning/test exports are also available under lerobot/ (see below). Some source XML/configuration fields retain original machine-specific asset paths. Re-execution requires the matching simulator and assets; compiled MuJoCo models are version-dependent. Consult per-episode files for actual observation modalities and recording details.

LeRobot v3 release

All 147 episodes are also available as 15 standalone LeRobot v3 datasets (five tasks × three splits), containing 2,094,957 synchronized frames at 50 Hz. Use separate navigation/wrist RGB views and named 11D robot states/actions. This RGB release preserves layout/configuration metadata; the raw archives retain models and states for depth reconstruction. Raw archives remain unchanged.

Select lerobot/<task>/fine_tuning, test_two_rooms, or test_multi_room as the local dataset root after downloading. The repository root is not one combined LeRobot dataset. Decoder tolerance is 1ms to accommodate float32 timestamp rounding in long videos, without changing the 20ms frame interval.

Only fine_tuning is for model fitting. Source demonstrations succeeded, but 50 Hz sample-and-hold physical replay has not been requalified; this distinction is retained in each dataset's provenance metadata. Read-back validation checks format, endpoint image decoding, counts and state/action dimensions, not learned policy success.

Downloads last month: -

Total file size:

83.6 GB