Datasets:

lemmingsandlemurs
/

piv0-iter0-success-pools

Tasks:

Robotics

Size:

Tags:

Dataset card Files Files and versions

xet

Community

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/hdf5/hdf5.py", line 49, in _split_generators
                  import h5py
              ModuleNotFoundError: No module named 'h5py'
              
              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 66, 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.

piv0 iter-0 rejection-success pools

Iter-0 rejection-sampled successful rollouts from three pi0-derived policies (lineages A2, B2, B3) across the active 12-goal packing-task suite.

These pools were built by:

Running each lineage's iter-0 policy in the dual-xArm packing environment.
Filtering rollouts by episode_metadata.success == True (the env's task-success flag).
Hardlinking the first N (sorted by (task_id, episode_id)) into a contiguous episode_0..N-1.hdf5 pool.

Every episode here is a confirmed task success. Pools are intended for filtered-BC / iterative bootstrapping (iter-1+).

Layout

iter-0/
  A2/{11 goals}/episode_0..39.hdf5
  B2/{11 goals}/episode_0..39.hdf5
  B3/{3 goals }/episode_0..39.hdf5

1000 episodes total (25 pools × 40 episodes).

Coverage matrix

✓ = 40-episode pool present · — = intentionally not in scope · ✗ = pool not built

Goal	Class	A2	B2	B3
SP-01	ID	✓	✓	—
SP-10	ID	✓	✓	—
SP-C01	ID	✓	✓	—
SP-D01	ID	✓	✓	—
SP-D03	ID	✗	✗	✗
SP-S01	ID	✓	✓	—
SP-B1	OOD	✓	✓	✓
SP-B2	OOD	✓	✓	✓
SP-B3	OOD	✓	✓	(24/40 partial — not uploaded)
SP-B7	OOD	✓	✓	—
SP-B9	OOD	✓	✓	(39/40 partial — not uploaded)
SP-B11	OOD	✓	✓	✓
per-lineage		11/12	11/12	3/12

Notes on the gaps:

SP-D03 has no iter-0 success pool for any lineage. Data collection is ongoing.
B3 iter-0 was OOD-focused by design; ID goals were not collected for B3 at iter-0.
B3/SP-B3 and B3/SP-B9 are excluded as partial pools (24/40 and 39/40 respectively, because the upstream rejection-sampling run produced fewer raw successes than the target N=40). They may be added later as a separate iter-0-partial/ subtree if needed.

File format

Each episode_*.hdf5:

attrs/episode_metadata (JSON): includes success: true and rollout context.
Datasets: per-step observations (multi-cam RGB, proprio), 14-D end-effector actions, and aux fields. Shapes vary by episode length.
Action convention used downstream: actions are zero-padded to 40-D for pi0 inference. Bytes [0:14] = leading pad, [14:28] = real 14-D actions, [28:40] = trailing pad (real_action_dim=14, real_action_start=14). Pools here store the real 14-D actions in their native format; the padding is applied at model boundary, not at storage.

Provenance

Pools were curated by build_piv0_success_pool.py from raw rollout trees produced by the iter-0 collection runs of each lineage (A2/B2/B3 = the three pi0 LoRA fine-tunes that define this generation of the bootstrap loop). The lineage labels track the experimental arm; consult the originating workspace for the precise checkpoint each pool sampled from.

License

Released under the terms applicable to the originating research workspace. Contact the dataset owner for clarification.

Downloads last month: 555

Total file size:

297 GB