The dataset viewer is not available for this subset.
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 80, in _split_generators
raise ValueError(
...<2 lines>...
)
ValueError: The TAR archives of the dataset should be in WebDataset format, but the files in the archive don't share the same prefix or the same types.
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.
SO-101 bimanual MuJoCo demonstrations (π0.7 recipe experiments)
Scripted-teacher and DAgger demonstrations for two SO-101 arms in MuJoCo, used to train the π0.7-recipe policies in
KuanKuanKuan/so101-pi07-sim-models. Three cameras (overhead, left wrist, right wrist), 50 Hz joint control.
Tasks: pick_place_any (one arm, cube to a disk), bin, handover (pass over a pad, then bin), two_bins
(one cube per arm into a shared bin), plus the older single-arm pick_place (sim_v1).
| folder | episodes | what |
|---|---|---|
sim_v1 |
800 (730 successful) | single-arm pick_place, scripted (first dataset) |
sim_ladder_v1 |
4000 (3509 successful) | four-task ladder, clean scripted teacher |
sim_recover_v1 |
2000 (1481 successful) | ladder with target knocks (recover style) |
sim_recover2_v1 |
5994 (4267 successful) | knocks, object bumps and deliberate missed grasps, mistake-labelled (recover2) |
sim_twobins_mixed_v1 |
1500 (1104 successful) | two_bins with either arm first (mixed_order and mixed_order_recover2) |
sim_dagger_v1 |
532 (430 successful) | DAgger from pi05_v1 step 8000 (broken takeover) |
sim_dagger_v2 |
570 (171 successful) | DAgger from pi05_2arm_v2 (broken takeover) |
sim_dagger_v3 |
546 (470 successful) | DAgger from pi05_warm2arm step 4000 (broken takeover) |
sim_dagger_v4 |
436 (207 successful) | DAgger from pi05_v1 step 15000 (broken takeover) |
sim_dagger_v5 |
293 (210 successful) | DAgger, two_bins only, from pi05_v1 step 17000 (broken takeover) |
sim_dagger_v6 |
329 (157 successful) | DAgger from pi05_2arm_v2 step 12000 (broken takeover) |
sim_dagger_v7 |
420 (362 successful) | DAgger from pi05_v1 step 20000, fixed takeover |
sim_dagger_v9 |
307 (265 successful) | DAgger from pi05_v1 step 20000, fixed takeover, new seeds |
sim_dagger_v10 |
424 (368 successful) | DAgger from pi05_dagger1 step 7000, fixed takeover |
Restore a folder: cat NAME/NAME.tar.part-* | tar xf - → NAME/episode_*/{meta.json, steps.npz, frames/<camera>/<step>.jpg}.
meta.json holds the instruction, task, style, success, quality (1–5) and the subtask names. steps.npz holds
per-step state, action (commanded joint targets, degrees and gripper 0–100), subtask index, mistake flags, and
object, goal and end-effector poses. Frames are 448 px JPEGs every frame_stride (5) steps. Failed episodes are
kept on purpose (π0.7 conditions on episode quality); PI_TRIM_STILL_TAIL_S=1 in the training code trims their
parked tails. Training mixes are symlink folders: mixes/*.txt lists their members, and sim_mix_v1_pi07_meta/ holds
the normalisation statistics and FAST tokenizer used by every model. The DAgger sets v1–v6 came from a teacher
takeover that was broken (fixed before v7; see the research log), so filter on success for those.
- Downloads last month
- 98