The dataset viewer is not available for this split.
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
text: int64
done_upto: int64
fit_ok: int64
fit_fail: int64
orientation: string
res: int64
to
{'done_upto': Value('int64'), 'res': Value('int64'), 'orientation': Value('string'), 'fit_ok': Value('int64'), 'fit_fail': Value('int64')}
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
text: int64
done_upto: int64
fit_ok: int64
fit_fail: int64
orientation: string
res: int64
to
{'done_upto': Value('int64'), 'res': Value('int64'), 'orientation': Value('string'), 'fit_ok': Value('int64'), 'fit_fail': Value('int64')}
because column names don't matchNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
LIBERO-PoseShift
Benchmark assets of LIBERO-PoseShift and of WACO (world-anchored canonicalization), from the paper Towards Pose-Robust Few-Shot Vision-Language-Action Models.
Project page · Code · Model
Wenhao Li, Zhongyi Han, Yuhao Wei, Liang Han, Yilong Yin, Liqiang Nie (Shandong University · Shenzhen Loop Area Institute · Harbin Institute of Technology, Shenzhen · Xidian University)
The benchmark
LIBERO-PoseShift is the first benchmark for RGB VLAs with graded camera, base and compound shifts. It extends LIBERO, whose tasks come with 50 demonstrations collected from one camera pose and one base pose, with a deterministic perturbation layer on its robosuite/MuJoCo environment, so that all methods see identical scenes.
- Camera shifts rotate and translate the third-person camera.
- Base shifts move the robot in the floor plane and rotate it about the vertical, leaving table, objects and third-person camera fixed.
- Both applies the two independently.
| Level | Camera translation | Camera rotation / axis | Base translation | Base yaw |
|---|---|---|---|---|
| L1 | ≤ 2 cm | ± 3° | ≤ 2 cm | ± 3° |
| L2 | ≤ 5 cm | ± 5° | ≤ 5 cm | ± 8° |
| L3 | ≤ 10 cm | ± 10° | ≤ 10 cm | ± 15° |
State and actions follow a base-frame protocol anchored at the calibration pose, in which a moved robot reports the pose it believes it has, exactly as hardware does. The main grid is LIBERO-Spatial and LIBERO-10 × 10 tasks × 10 conditions (clean plus three kinds at three levels) × 10 episodes, 2,000 episodes per model.
The perturbation layer itself is code: it is registered with LeRobot as --env.type=libero_poseshift by the
code repository. This dataset holds everything else that is needed to run the
benchmark with the WACO runtime and to reproduce the error-aware fine-tuning. Training demonstrations are the
public LIBERO demonstrations in LeRobot format (lerobot/libero),
used without modification.
Contents
| Path | Size | Content |
|---|---|---|
poseshift_perturbations.json |
0.6 MB | the sampled shift of every (task, episode, kind, level) of the four LIBERO suites, as drawn by the benchmark's deterministic sampler (reference for other implementations) |
calib_libero/<suite>__<task>.npz |
4.7 MB | calibration bundle of each of the 40 LIBERO tasks |
calib_liberoplus/<suite>__<task>.npz |
81 MB | calibration bundles recorded in the LIBERO-Plus scenes, with a twelve-view scan of the empty workspace |
depth_dav2_128.npy, depth_dav2_128.json |
8.4 GB | per-frame metric depth of the 273,465 third-person frames of lerobot/libero, for the round-trip reprojection of the error-aware fine-tuning |
fewshot_episodes_10.json, fewshot_episodes_20.json |
< 1 MB | episode indices of the first 10 / 20 demonstrations of every task (few-shot scaling) |
liberoplus_camera_ids.json |
< 1 MB | task ids of the LIBERO-Plus camera track per suite (1,599 variants) |
Calibration bundles
A calibration bundle is recorded once per workspace and is the only calibration WACO assumes; on a real robot it is a
photograph of the bare table taken during setup. Every .npz contains
| Key | Shape | Meaning |
|---|---|---|
ref_rgb |
(256, 256, 3) uint8 | RGB frame of the empty workspace from the nominal camera (raw MuJoCo orientation, row 0 at the bottom) |
ref_depth |
(256, 256) float32 | metric depth of the same frame (metres along the optical axis) |
ref_mask |
(256, 256) bool | static-scene mask (projected arm and invalid depth removed) |
T_wb0 |
(4, 4) float64 | nominal base pose in the world frame |
T_bc0 |
(4, 4) float64 | camera-to-base extrinsic, measured once before data collection |
f_px |
scalar | focal length in pixels |
language |
string | task instruction |
The LIBERO-Plus bundles additionally contain the scan views_T (12, 4, 4), views_rgb, views_depth, views_mask
and views_kind: twelve further empty-workspace views (azimuth ±20°, ±45°, ±70° at the nominal height, ±45°, ±70°
raised by 15°, a view 1.6× farther away, and one raised view), each with its pose in the nominal camera frame.
Per-frame depth
depth_dav2_128.npy is a float16 array of shape (273465, 128, 128), indexed by the global frame index of
lerobot/libero, in the orientation of the dataset's images. It is predicted by Depth Anything V2 and aligned per
frame to the calibration scale on the static background of the reference frame. Open it memory-mapped:
import numpy as np
depth = np.load("depth_dav2_128.npy", mmap_mode="r") # metres
Usage
git clone https://github.com/peacelwh/WACO.git && cd WACO
bash scripts/download_assets.sh # everything except the depth -> data/
WITH_DEPTH=1 bash scripts/download_assets.sh # + per-frame depth (training only)
or directly
hf download Bluexxx/LIBERO-PoseShift --repo-type dataset --local-dir data --exclude depth_dav2_128.npy
All files can be regenerated from the simulator with the scripts of the code repository
(build_calibration.py, build_calibration_liberoplus.py, precompute_depth.py, fewshot_episodes.py,
liberoplus_camera_ids.py).
License
Released under Apache-2.0. The calibration frames are rendered from the scenes of
LIBERO and
LIBERO-Plus, and the depth is derived from the frames of
lerobot/libero; please also follow their licenses.
Citation
@article{li2026waco,
title = {Towards Pose-Robust Few-Shot Vision-Language-Action Models},
author = {Li, Wenhao and Han, Zhongyi and Wei, Yuhao and Han, Liang and Yin, Yilong and Nie, Liqiang},
journal = {arXiv preprint},
year = {2026}
}
- Downloads last month
- 81