Dataset Viewer
The dataset viewer is not available for this subset.
Cannot get the split names for the config 'default' of the dataset.
Exception:    SplitsNotFoundError
Message:      The split names could not be parsed from the dataset config.
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 286, in get_dataset_config_info
                  for split_generator in builder._split_generators(
                                         ~~~~~~~~~~~~~~~~~~~~~~~~~^
                      StreamingDownloadManager(base_path=builder.base_path, download_config=download_config)
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  )
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/webdataset/webdataset.py", line 88, in _split_generators
                  inferred_arrow_schema = pa.concat_tables(pa_tables, promote_options="default").schema
                                          ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "pyarrow/table.pxi", line 6268, in pyarrow.lib.concat_tables
                File "pyarrow/error.pxi", line 155, in pyarrow.lib.pyarrow_internal_check_status
                File "pyarrow/error.pxi", line 92, in pyarrow.lib.check_status
                  raise convert_status(status)
              pyarrow.lib.ArrowTypeError: Unable to merge: Field json has incompatible types: struct<attributes: struct<embodiment: string, robot: string, env: string, task_name: string, task_description: string, fps: int64, total_frames: int64, source: string, speed_meta_done: bool, mimicdroid_play: bool, subgoals: string, n_subgoals: int64, action57_done: bool, state57_abs: bool, dim: int64, sidecams_done: bool>, zarr_format: int64, node_type: string, shape: list<item: int64>, data_type: string, chunk_grid: struct<name: string, configuration: struct<chunk_shape: list<item: int64>>>, chunk_key_encoding: struct<name: string, configuration: struct<separator: string>>, fill_value: double, codecs: list<item: struct<name: string, configuration: struct<endian: string, level: int64, checksum: bool>>>, storage_transformers: list<item: null>> vs struct<shape: list<item: int64>, data_type: string, chunk_grid: struct<name: string, configuration: struct<chunk_shape: list<item: int64>>>, chunk_key_encoding: struct<name: string, configuration: struct<separator: string>>, fill_value: bool, codecs: list<item: struct<name: string, configuration: struct<level: int64, checksum: bool>>>, attributes: struct<>, zarr_format: int64, node_type: string, storage_transformers: list<item: null>>: Unable to merge: Field fill_value has incompatible types: double vs bool
              
              The above exception was the direct cause of the following exception:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/config/split_names.py", line 68, in compute_split_names_from_streaming_response
                  for split in get_dataset_split_names(
                               ~~~~~~~~~~~~~~~~~~~~~~~^
                      path=dataset,
                      ^^^^^^^^^^^^^
                      config_name=config,
                      ^^^^^^^^^^^^^^^^^^^
                      token=hf_token,
                      ^^^^^^^^^^^^^^^
                  )
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 340, in get_dataset_split_names
                  info = get_dataset_config_info(
                      path,
                  ...<6 lines>...
                      **config_kwargs,
                  )
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 291, in get_dataset_config_info
                  raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
              datasets.inspect.SplitsNotFoundError: The split names could not be parsed from the dataset config.

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.

egoverse-v10-state

The EgoVerse v10 pre-training set: 13,469 egocentric episodes in the 57-d state/action format, plus the retrieval manifest that pairs them. This is the set the one-shot policies are pre-trained on before any RoboCasa / Tabletop post-training.

State and action only — no video. Each episode in the source dataset is ~352 MB, of which ~348 MB is three RGB views; dropping them takes the set from 5.3 TB to 46.8 GB and keeps everything the state-space pre-training recipe reads.

count
episodes 13,469 (MECKA 6,718, scale 6,074, aria 657, mimicdroid 20)
windows 1,092,145 (1,058,666 kept)
queries 945,898, top-20 matches each
window length 90 frames (egocentric human), 60 frames (mimicdroid)

eva episodes (668) are excluded: they contribute 0 queries and 0 matches in the v10 matching, so they are never read during pre-training. Their windows are still listed in manifest/windows.npz, which is kept byte-identical to the one used for training.

Layout

data/<episode>.tar      # 9,900 episodes
data2/<episode>.tar     # 3,569 episodes  <- HF allows at most 10,000 files per directory
manifest/               # retrieval manifest + normalization statistics
  windows.npz           #   ep / w0 / w1 / isev / emb / cls / keep / source per window
  pairs.npz             #   raw matching
  pairs_clean.npz       #   query_idx, match_idx (N, 20), match_dist, match_speed  <- used by training
  query_eps_q{10,25,50,75}.json   #   query-episode subsets (the q10 / q25 / … ablations)
  holdout_eps.txt       #   episodes held out of matching
  state57_stats*.json   #   state normalization (`_pctl` = percentile-hybrid, the v10 default)
  action57_stats.json
  t5_embeddings.pkl     #   per-episode T5 text embeddings
episodes.txt            # the 13,469 episode names
stats.json              # the table above, machine-readable

Unpack both directories into one root; data/ and data2/ are a storage artifact, not a split.

Each tar unpacks to a zarr group:

<episode>/
  zarr.json
  state57/{state, valid}        # (T, 57)
  action57/{action, valid}      # (T, 57)
  meta/{eef_speed, ...}         # per-frame metadata used by the motion matching

Four episodes (69b932ae36bea964e3c3ae67, 69bad8c0e423ad04bb1707c5, 69baf1907236446f4f672422, 69bb01723abe312e772bec7b) predate that layout: they carry action57, annotations, left/right.obs_* and obs_head_pose* instead of meta/ and state57/, and are shipped as-is.

Use

Unpack to the EgoVerse root the training config expects and point it at the manifest:

HUM_ROOT = <...>/data          # episode names in episodes.txt / windows.npz["ep"]
manifest = <...>/manifest      # windows.npz + pairs_clean.npz + state57_stats_pctl.json

Episode names are the manifest's ep field verbatim, so no renaming is needed.

Downloads last month
28