Hugging Face's logo

Datasets:
Yinpei
/
RoboMME-plus-model-rollout

Modalities:
Video
Languages:
English
Size:
1K<n<10K
Tags:
robotics
robomme
rollouts
video
trajectories
Dataset card Files
xet
 Community
Dataset Viewer
Duplicate
The dataset viewer is not available for this split.
Cannot load the dataset split (in streaming mode) to extract the first rows.
Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
schema_version: int64
created_utc: string
selection: string
excluded_methods: list<item: string>
  child 0, item: string
episodes: int64
methods: struct<embodied-astra: struct<episodes: int64, counts: struct<success: int64, fail: int64, timeout:  (... 9172 chars omitted)
  child 0, embodied-astra: struct<episodes: int64, counts: struct<success: int64, fail: int64, timeout: int64>, success_rate_re (... 1480 chars omitted)
      child 0, episodes: int64
      child 1, counts: struct<success: int64, fail: int64, timeout: int64>
          child 0, success: int64
          child 1, fail: int64
          child 2, timeout: int64
      child 2, success_rate_recorded: double
      child 3, per_task: struct<BinFill: struct<episodes: int64, counts: struct<success: int64, fail: int64>>, PickXtimes: st (... 1353 chars omitted)
          child 0, BinFill: struct<episodes: int64, counts: struct<success: int64, fail: int64>>
              child 0, episodes: int64
              child 1, counts: struct<success: int64, fail: int64>
                  child 0, success: int64
                  child 1, fail: int64
          child 1, PickXtimes: struct<episodes: int64, counts: struct<success: int64, fail: int64>>
              child 0, episodes: int64
              child 1, counts: struct<success: int64, fail: int64>
                  child 0, success: int64
                  child 1, fail: int64
          child 2, SwingXtimes: struct<episodes: int64, counts: struct<success: int64, fail: int64>>
  
...
<item: int64>>
          child 0, PickHighlight: list<item: int64>
              child 0, item: int64
      child 1, embodied-astra-fs: struct<PickHighlight: list<item: int64>>
          child 0, PickHighlight: list<item: int64>
              child 0, item: int64
      child 2, framesamp-modul: struct<PickHighlight: list<item: int64>>
          child 0, PickHighlight: list<item: int64>
              child 0, item: int64
      child 3, simplememvla: struct<PickHighlight: list<item: int64>>
          child 0, PickHighlight: list<item: int64>
              child 0, item: int64
      child 4, 3-tier-astra: struct<PickHighlight: list<item: int64>, ButtonUnmaskSwap: list<item: int64>>
          child 0, PickHighlight: list<item: int64>
              child 0, item: int64
          child 1, ButtonUnmaskSwap: list<item: int64>
              child 0, item: int64
  child 4, unchanged_episodes: int64
repository: string
total_archive_bytes: int64
verification: string
archives: list<item: struct<method: string, filename: string, episodes: int64, files: int64, unpacked_bytes: i (... 63 chars omitted)
  child 0, item: struct<method: string, filename: string, episodes: int64, files: int64, unpacked_bytes: int64, bytes (... 51 chars omitted)
      child 0, method: string
      child 1, filename: string
      child 2, episodes: int64
      child 3, files: int64
      child 4, unpacked_bytes: int64
      child 5, bytes: int64
      child 6, sha256: string
      child 7, recorded_successes: int64
to
{'schema_version': Value('int64'), 'repository': Value('string'), 'episodes': Value('int64'), 'created_utc': Value('timestamp[s]'), 'total_archive_bytes': Value('int64'), 'archives': List({'method': Value('string'), 'filename': Value('string'), 'episodes': Value('int64'), 'files': Value('int64'), 'unpacked_bytes': Value('int64'), 'bytes': Value('int64'), 'sha256': Value('string'), 'recorded_successes': Value('int64')}), 'verification': 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
              schema_version: int64
              created_utc: string
              selection: string
              excluded_methods: list<item: string>
                child 0, item: string
              episodes: int64
              methods: struct<embodied-astra: struct<episodes: int64, counts: struct<success: int64, fail: int64, timeout:  (... 9172 chars omitted)
                child 0, embodied-astra: struct<episodes: int64, counts: struct<success: int64, fail: int64, timeout: int64>, success_rate_re (... 1480 chars omitted)
                    child 0, episodes: int64
                    child 1, counts: struct<success: int64, fail: int64, timeout: int64>
                        child 0, success: int64
                        child 1, fail: int64
                        child 2, timeout: int64
                    child 2, success_rate_recorded: double
                    child 3, per_task: struct<BinFill: struct<episodes: int64, counts: struct<success: int64, fail: int64>>, PickXtimes: st (... 1353 chars omitted)
                        child 0, BinFill: struct<episodes: int64, counts: struct<success: int64, fail: int64>>
                            child 0, episodes: int64
                            child 1, counts: struct<success: int64, fail: int64>
                                child 0, success: int64
                                child 1, fail: int64
                        child 1, PickXtimes: struct<episodes: int64, counts: struct<success: int64, fail: int64>>
                            child 0, episodes: int64
                            child 1, counts: struct<success: int64, fail: int64>
                                child 0, success: int64
                                child 1, fail: int64
                        child 2, SwingXtimes: struct<episodes: int64, counts: struct<success: int64, fail: int64>>
                
              ...
              <item: int64>>
                        child 0, PickHighlight: list<item: int64>
                            child 0, item: int64
                    child 1, embodied-astra-fs: struct<PickHighlight: list<item: int64>>
                        child 0, PickHighlight: list<item: int64>
                            child 0, item: int64
                    child 2, framesamp-modul: struct<PickHighlight: list<item: int64>>
                        child 0, PickHighlight: list<item: int64>
                            child 0, item: int64
                    child 3, simplememvla: struct<PickHighlight: list<item: int64>>
                        child 0, PickHighlight: list<item: int64>
                            child 0, item: int64
                    child 4, 3-tier-astra: struct<PickHighlight: list<item: int64>, ButtonUnmaskSwap: list<item: int64>>
                        child 0, PickHighlight: list<item: int64>
                            child 0, item: int64
                        child 1, ButtonUnmaskSwap: list<item: int64>
                            child 0, item: int64
                child 4, unchanged_episodes: int64
              repository: string
              total_archive_bytes: int64
              verification: string
              archives: list<item: struct<method: string, filename: string, episodes: int64, files: int64, unpacked_bytes: i (... 63 chars omitted)
                child 0, item: struct<method: string, filename: string, episodes: int64, files: int64, unpacked_bytes: int64, bytes (... 51 chars omitted)
                    child 0, method: string
                    child 1, filename: string
                    child 2, episodes: int64
                    child 3, files: int64
                    child 4, unpacked_bytes: int64
                    child 5, bytes: int64
                    child 6, sha256: string
                    child 7, recorded_successes: int64
              to
              {'schema_version': Value('int64'), 'repository': Value('string'), 'episodes': Value('int64'), 'created_utc': Value('timestamp[s]'), 'total_archive_bytes': Value('int64'), 'archives': List({'method': Value('string'), 'filename': Value('string'), 'episodes': Value('int64'), 'files': Value('int64'), 'unpacked_bytes': Value('int64'), 'bytes': Value('int64'), 'sha256': Value('string'), 'recorded_successes': Value('int64')}), 'verification': 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.

RoboMME-Plus Model Rollouts

4,800 episodes: six methods × 16 tasks × 50 episodes. Both OOD-fine-tuned FrameSamp runs are excluded. This collection contains recorded evaluations from multiple benchmark revisions; per-episode metadata and producing configurations identify the saved trials.

Method Episodes Recorded successes Recorded success rate
embodied-astra 800 181 22.625%
embodied-astra-fs 800 282 35.250%
framesamp-modul 800 22 2.750%
simplememvla 800 16 2.000%
3-tier-astra 800 76 9.500%
ponderpounce 800 23 2.875%

Each METHOD/TASK/episode_NNN/ contains exactly six files:

  • episode.mp4: original 20-FPS rollout, front camera left and wrist right; red-bordered demonstration frames are included where recorded.
  • actions.npy: original numeric actions, one row per environment call.
  • steps.jsonl.gz: recorded actions, state trajectories, rewards, terminal flags, environment info and trial-reset events.
  • frames.jsonl.gz: every video frame mapped to demo/execution phase and control-step/batch timing.
  • metadata.json: the exact episode identity and settings used in the saved rollout.
  • result.json: recorded outcome, task goal, timing, action conventions and producing configuration ID.

Original/one-shot Astra use 7D waypoint commands [x,y,z,roll,pitch,yaw,gripper]; the other methods use 8D joint targets [q1,...,q7,gripper]. Units are metres/radians; gripper commands use −1 closed and +1 open. Recorded gripper observations are finger widths in metres. Waypoint commands can span many control steps: align using step_before/step_after and the frame map. These are recorded evaluator commands and returned robot-state batches. Initial robot states are omitted when the original recorder did not save them. Trial reset records can add control steps without an action-array row.

manifest.json provides per-file SHA256 values and manifest.csv indexes all episodes. provenance/ holds deduplicated producing configurations and code/metadata hashes. benchmark_reference.json records benchmark reference snapshots; per-episode metadata describes each saved rollout. audit.json records artifact validation and its limits. Model-request logs, prompts, credentials, host paths, raw image caches, old attempts and duplicate rollouts are excluded.

Read compressed trajectories with gzip.open(path, "rt") and parse each line with json.loads. Load actions with numpy.load(path, allow_pickle=False). Verify all exported files with sha256sum -c checksums.sha256.

The collection includes the 2026-10-28 evaluations of all 50 PickHighlight episodes for original Astra, one-shot Astra, FrameSamp+Modul, SimpleMemVLA and 3-tier Astra, plus all 50 ButtonUnmaskSwap episodes for 3-tier Astra. The remaining 4,500 saved trials are preserved. One-shot Astra uses image-only training examples. ButtonUnmask and VideoUnmask test rollouts use the restored no-removal version.

Download and extract

One gzip-compressed tar archive is provided per model. Each archive contains its model directory, 800 episode directories and a model summary. Shared manifests, provenance and benchmark-reference metadata are provided as separate files at the repository root. Both OOD-fine-tuned FrameSamp variants are excluded.

Archive Episodes Compressed size Recorded successes
embodied-astra.tar.gz 800 1.68 GB 181
embodied-astra-fs.tar.gz 800 1.76 GB 282
framesamp-modul.tar.gz 800 1.92 GB 22
simplememvla.tar.gz 800 1.83 GB 16
3-tier-astra.tar.gz 800 2.06 GB 76
ponderpounce.tar.gz 800 2.06 GB 23

Download the dataset and verify the archives:

hf download Yinpei/RoboMME-plus-model-rollout --repo-type dataset --local-dir RoboMME-Plus-model-rollout
cd RoboMME-Plus-model-rollout
sha256sum -c archives.sha256
for archive in *.tar.gz; do tar -xzf "$archive"; done
sha256sum -c checksums.sha256

To extract a single model, run tar -xzf embodied-astra.tar.gz, for example. archives.json records exact compressed sizes, archive hashes and file counts. checksums.sha256 verifies the extracted rollout files and shared metadata. The archives preserve the saved videos and numeric trajectories, including the 300 model trials evaluated on 2026-10-28.
Downloads last month
90
Total file size:
11.3 GB

Collection including Yinpei/RoboMME-plus-model-rollout