Datasets:

RoMALab
/

RoboTwin2.0-15tasks-randomized500-wide

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 81, in _split_generators
                  first_examples = list(islice(pipeline, self.NUM_EXAMPLES_FOR_FEATURES_INFERENCE))
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/webdataset/webdataset.py", line 32, in _get_pipeline_from_tar
                  fs: fsspec.AbstractFileSystem = fsspec.filesystem("memory")
                                                  ~~~~~~~~~~~~~~~~~^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/fsspec/registry.py", line 302, in filesystem
                  cls = get_filesystem_class(protocol)
                File "/usr/local/lib/python3.14/site-packages/fsspec/registry.py", line 239, in get_filesystem_class
                  raise ValueError(f"Protocol not known: {protocol}")
              ValueError: Protocol not known: memory
              
              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.

RoboTwin 2.0 randomized-500 wide-view rerenders

This is a derived, wide-camera rerender of 7,379 complete RoboTwin 2.0 episodes across 15 bimanual manipulation tasks. It contains 1,530,938 JPEG frames plus per-episode NumPy metadata and actions.

The source conversion produced metadata for 7,500 candidate episodes, but 121 rerenders had no output frames and are deliberately excluded. The exact task-local episode indices are listed in manifest.json.

The source trajectories come from TianxingChen/RoboTwin2.0. Rerendering and conversion used video2vla.

Format

The dataset uses 15 uncompressed task archives because the source tree contains more than 1.5 million small files. Each archive contains both its matching episodes/*.npz files and frames/<task>/... tree. Extract all archives into one directory to recreate the dataset root:

huggingface-cli download RoMALab/RoboTwin2.0-15tasks-randomized500-wide --repo-type dataset --local-dir robotwin-wide-hub
mkdir -p robotwin-wide
for archive in robotwin-wide-hub/archives/*.tar; do
  tar -xf "$archive" -C robotwin-wide
done

Each episode NPZ has these fields:

episode_id, task, embodiment, subset, and language
actions: float32 array shaped [T, 14] with absolute joint actions
image_paths: T relative paths into frames/
action_type: joint
image_paths_are_placeholders: False

Frames are 640×352 JPEG images rendered from the head camera with a 60° field of view. Episode lengths are variable. No train/validation split is imposed.

Contents and checksums

Task	Episodes	Frames	Archive	SHA-256
`adjust_bottle`	500	71,723	`archives/adjust_bottle.tar`	`aa0449a9f8feeff1ab7dab8cabe953c026b2fe62d7e1d89d9c789a6cb2ae4c34`
`beat_block_hammer`	498	57,470	`archives/beat_block_hammer.tar`	`fd61fb5b761315f3c49b15b4c3266ee84a236c8f9ae32cd30d639ff802a1ea6e`
`click_alarmclock`	500	42,767	`archives/click_alarmclock.tar`	`2a583b7c81cde25dd718e50408221ef8713415a0763bc43dd9211bdea7f436cb`
`dump_bin_bigbin`	500	118,637	`archives/dump_bin_bigbin.tar`	`7aa3111ea2159beaa5cfbb8d9dacf25587f732f7091014794c224216d376fe6f`
`grab_roller`	500	48,321	`archives/grab_roller.tar`	`ce4c52618619aae8aeb56719492893bea8e1567c236d252b6ecdcab4cddae737`
`handover_block`	500	142,746	`archives/handover_block.tar`	`ce185314dcb9101e36ffe02563088fc53156a368478891781f3bee649f79617d`
`lift_pot`	500	56,678	`archives/lift_pot.tar`	`35f8e8f6a886a5f626f21979c7fb674ce87d5a9b4c250c8fd5f445312c391f2b`
`open_laptop`	500	111,893	`archives/open_laptop.tar`	`0986ebfd3aa2f7a8749637a84dbd6724fcfe27a8f75894176c8d359ea3b908da`
`open_microwave`	500	276,044	`archives/open_microwave.tar`	`a21218686ca3b238b45246e2b604ff54b4ffb66754711bc959983413bf69aa87`
`place_cans_plasticbox`	500	143,203	`archives/place_cans_plasticbox.tar`	`d2aa2de5f61d0e89e3b38552b706ddbd377e0b5634087d4be49b2fc0e0610293`
`place_phone_stand`	500	64,024	`archives/place_phone_stand.tar`	`f0a7bbf85824e97d4f5d9730a2ab66f48f71bb355818b0c7ca044d039340c9a2`
`put_object_cabinet`	403	109,809	`archives/put_object_cabinet.tar`	`0e7fb20f56c7dcfca6eab84b94493221ad63535fddf781daa681648cdd5846a1`
`scan_object`	478	80,908	`archives/scan_object.tar`	`6b6fdda10b2fa086481c0e6f713b1f0883cb02aeca51d02b88b6887a67f89c1c`
`stack_blocks_two`	500	156,880	`archives/stack_blocks_two.tar`	`134071938e295b591d5c5510d114533e778a431cf1f53feef503a6eb3920077f`
`turn_switch`	500	49,835	`archives/turn_switch.tar`	`3f196fe0e90e2ccd6a1ee9a752ff1e6fd762658bc5bfc81f627aab8e3fbf3eef`

Machine-readable counts, byte sizes, render settings, and checksums are in manifest.json. The generation configuration is included under generation/.

Loading one episode

from pathlib import Path
import numpy as np

root = Path("robotwin-wide")
episode_path = next((root / "episodes").glob("*.npz"))
with np.load(episode_path, allow_pickle=True) as episode:
    actions = episode["actions"]
    frame_paths = [root / str(path) for path in episode["image_paths"]]

Lineage and license

This dataset is derived from RoboTwin 2.0 and retains its MIT license. See the RoboTwin 2.0 paper, project page, and source code.

Downloads last month: 145

Total file size:

73.4 GB

Paper for RoMALab/RoboTwin2.0-15tasks-randomized500-wide

RoboTwin 2.0: A Scalable Data Generator and Benchmark with Strong Domain Randomization for Robust Bimanual Robotic Manipulation

Paper • 2506.18088 • Published Jun 22, 2025 • 19