Hugging Face's logo

Datasets:
NeoteAI
/
OpenNeoData-OXT

Languages:
English
Tags:
robotics
tactile
zarr
open-x-tactile
License:
Dataset card Files
xet
 Community
1

You need to agree to share your Hugging Face information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

OpenNeoData-OXT

Corrected pilot and verified bulk data. Source: NeoteAIEmbodied/OpenNeoData at e9c6eb33c6c2971c39edbbbcc135365b90d0dd22.

Conversion profile: oxt_organizer_20261024_v1. UR wrist rotations are right-multiplied by local Rz(+90°) exactly once and stored as rotation vectors. Translations, joints, gripper values, timestamps and original source records are unchanged. Both UMI wrists retain their original stored poses and shared-coordinate transformation.

Camera RGB and visual tactile images use lossy per-frame JPEG, quality 90, 4:4:4, RGB uint8, native resolution. Camera chunks contain one frame; tactile chunks contain one frame and one sensor. Numerical arrays remain lossless. This representation changes decoded image pixels; source hashes and stored-output hashes are recorded separately in the external audit bundle.

The conservative internal quality-review triggers are PSNR below 35 dB or SSIM below 0.98. These are not OXT standards and do not override visible-detail concerns or prior review holds.

Data conventions and retained limitations

Each archive is self-contained Zarr format 2: camera RGB arrays have shape (T,H,W,3) and paired visual tactile arrays have shape (T,2,H,W,3). Complete episodes are retained without new trimming, interpolation or recording repairs. The JPEG conversion adds lossy image compression to the source videos' existing compression.

Measured state, timestamps, episode boundaries and instructions are retained. Source column-based rot6d is converted to rotvec where wrist poses are exported. UMI shared-frame trajectories retain the matched original curation chain; no UMI arm joints are invented. Fixed dual-arm wrist poses remain omitted where inter-base calibration is unavailable. Gripper opening uses slot 28; InTacS1 areas 0/1 denote the source left/right tactile streams within each gripper. FTP-1 derives training actions from state trajectories.

Original review findings and contributor decisions remain in the external audit evidence. This revision does not claim that every raw VLM page passed or that new organizer acceptance has been granted. Visual review does not establish calibrated force, physical capture-clock accuracy or quantitative hardware latency. UMI previews cover the documented excerpts only; stored episodes remain complete.

Loading

Images are already compressed inside Zarr v2 chunks: JPEG quality 90, 4:4:4, native-resolution RGB uint8. This is lossy image storage; numerical arrays remain lossless. Changing the reader does not re-encode the archive.

Use the official OXT ReplayBuffer from commit 5a587afba3e6c142dd287c7eaa35f0cbe017cd7f (October 30 compression update). Importing it registers the standard image codecs automatically; no custom JPEG registration is needed. The verified reader environment is Python 3.11 with:

python -m pip install numpy==2.2.6 zarr==3.1.3 numcodecs==0.16.3 imagecodecs==2025.3.30

Unzip an archive, add the official checkout's OXT-QC directory to PYTHONPATH, and open the inner dataset directory:

from common.replay_buffer import ReplayBuffer

buffer = ReplayBuffer.create_from_path("OpenNeoData-OXT/<platform>/<dataset>.zarr", mode="r")
# Bound memory use when an episode contains many full-resolution image streams.
start, end = buffer.get_episode_slice(0).start, buffer.get_episode_slice(0).stop
for first in range(int(start), int(end), 8):
    frames = buffer.get_steps_slice(first, min(int(end), first + 8))
    # Consume this window before reading the next one.
# buffer.get_episode(0) also works when the complete episode fits in memory.

The separate, unchanged FTP-1 ZarrDataset loader at commit 89fa681d6c014cce28300946b7526db808e0b1c1 remains supported in its verified environment. Keep Python runtimes and their binary packages separate.

QC profile oxt_qc_20261030_v1 verifies actual compression metadata and full image-stream decoding with the official reader. OXT's recommended default for new writes is Blosc-Zstd level 5 with BITSHUFFLE; readers use each array's stored codec, so that default does not require re-encoding compatible JPEG archives. Compatibility verification is not a claim of new organizer acceptance.

Revision and audit evidence

This revision replaces the 15 already-verified episodes (8,223 frames) from baseline commit ed333a41edc72461e6caf9a5700b8f460e4bafe2. The baseline remains in repository history. It does not approve held episodes. QC reports, certificates, conversion manifests, review videos and the corrected source-axis UMI previews are kept in the owned external audit workspace and separate organizer bundle. The public branch contains data ZIPs, this README and required .gitattributes only.

License and attribution

Dataset: CC BY-NC-SA 4.0. Attribution: NeoteAI / OpenNeoData. Conversion changes are described above; upstream tools retain their respective licenses. Contributor contact: yangtianqwer@gmail.com.

Original release tools: OXT-QC at cadbf5e7d1cfcec861a6d58779fc6d6e057f3126, and FTP-1 at 89fa681d6c014cce28300946b7526db808e0b1c1.

Archives

Each ZIP contains only OpenNeoData-OXT/<platform>/<dataset>.zarr/.... The following machine-readable listing gives exact archive bytes, SHA-256, profile, episode IDs and frame counts. Links use the current revision; reproducible consumers should pin the verified commit.

{ "aloha__task_0015_test_tube_cleaning_and_collection__episode_11118.zarr.zip": { "bytes": 434809955, "conversion_profile": "oxt_organizer_20261024_v1", "episodes": [ { "episode_index": 11118, "frames": 1454, "platform": "aloha" } ], "sha256": "e7da16ecd68e823938d12bd48930f4e32cf201dcc24b73a63cf8a984d3e4dd28" }, "arx5__task_0009_carlights_pick_and_insert.zarr.zip": { "bytes": 113391301, "conversion_profile": "oxt_organizer_20261024_v1", "episodes": [ { "episode_index": 2345, "frames": 301, "platform": "arx5" } ], "sha256": "311812988c79466d627598eb7a6af0f5f4f09fdf954f53988850ebc3079df69d" }, "arx5_single__task_0003_clip_chips.zarr.zip": { "bytes": 63495724, "conversion_profile": "oxt_organizer_20261024_v1", "episodes": [ { "episode_index": 436, "frames": 305, "platform": "arx5_single" } ], "sha256": "98af987fed924b6d98fb73ec9f7dcc301c8e5429eb72c021ee30d3eb2679715c" }, "bulk/3e3b5139e7b3602b957efd5bafd0c1a5c0bcaf027b2c0476d706593777ad0632/51ba4a8a61168f9a5301/ur__task_0047__part_0000.zarr.zip": { "bytes": 411534730, "conversion_profile": "oxt_organizer_20261024_v1", "episodes": [ { "episode_index": 16555, "frames": 424, "platform": "ur" }, { "episode_index": 16556, "frames": 460, "platform": "ur" }, { "episode_index": 16558, "frames": 374, "platform": "ur" }, { "episode_index": 16559, "frames": 389, "platform": "ur" }, { "episode_index": 16560, "frames": 393, "platform": "ur" } ], "sha256": "beabd4c1fe680ff022ff441aaa09cf0a3b8d8c7d9274ac7f5577f6a4d4afb76f" }, "bulk/3e3b5139e7b3602b957efd5bafd0c1a5c0bcaf027b2c0476d706593777ad0632/8e1f35ca0728f8c8d692/ur__task_0047__part_0000.zarr.zip": { "bytes": 175298599, "conversion_profile": "oxt_organizer_20261024_v1", "episodes": [ { "episode_index": 16553, "frames": 425, "platform": "ur" }, { "episode_index": 16554, "frames": 447, "platform": "ur" } ], "sha256": "f41599fa66b2c29949350ea58a4ea2510c0c09f5113419bf7730a63f679cc608" }, "bulk/7f789cfcbf6f20bc2902a77f3467873d6bf3b0b0db56c4804600574bf1297083/1588c8bf8dd80cc87920/ur__task_0047__part_0000.zarr.zip": { "bytes": 69794594, "conversion_profile": "oxt_organizer_20261024_v1", "episodes": [ { "episode_index": 16621, "frames": 338, "platform": "ur" } ], "sha256": "f14198acd1bf0f1b06eccdae4fa9827f466fafbf34c17206f122023cf35663e1" }, "bulk/dfe17b165f1f7785f480f03385d650c996db602525e75a07112af845e7b058f0/a09f038225e40d0d8bb7/ur__task_0047__part_0000.zarr.zip": { "bytes": 69950491, "conversion_profile": "oxt_organizer_20261024_v1", "episodes": [ { "episode_index": 16582, "frames": 346, "platform": "ur" } ], "sha256": "d12262533f9acff6555e8d35cab5bdf654e427d973bf0c6af2ff06643efd0138" }, "bulk/dfe17b165f1f7785f480f03385d650c996db602525e75a07112af845e7b058f0/ac6173f6c99ec3c373ba/ur__task_0047__part_0000.zarr.zip": { "bytes": 77946379, "conversion_profile": "oxt_organizer_20261024_v1", "episodes": [ { "episode_index": 16583, "frames": 385, "platform": "ur" } ], "sha256": "9bbe7fb4aead7026ed8be892385cd2c1e7511e0161e49ecdfee6df6cf8aa1d91" }, "bulk/dfe17b165f1f7785f480f03385d650c996db602525e75a07112af845e7b058f0/daf175ccfd067791d107/ur__task_0047__part_0000.zarr.zip": { "bytes": 87377939, "conversion_profile": "oxt_organizer_20261024_v1", "episodes": [ { "episode_index": 16586, "frames": 412, "platform": "ur" } ], "sha256": "8477cd9e21b6d883e27977f87b99ebc411db223389ac294146056660bacf83a7" }, "bulk/f86a19985d5904bd3859e2883e445c81c7b3d87a2fcece936116b66a2a685df8/2ebded60c4c2c1f07e9d/ur__task_0047__part_0000.zarr.zip": { "bytes": 64813299, "conversion_profile": "oxt_organizer_20261024_v1", "episodes": [ { "episode_index": 16566, "frames": 321, "platform": "ur" } ], "sha256": "596429de026941d5d170fc50afa361c11cef77ce58d768c439e5de52ee4405e3" }, "bulk/f86a19985d5904bd3859e2883e445c81c7b3d87a2fcece936116b66a2a685df8/4baf76b52acfa8e3d901/ur__task_0047__part_0000.zarr.zip": { "bytes": 70062833, "conversion_profile": "oxt_organizer_20261024_v1", "episodes": [ { "episode_index": 16562, "frames": 344, "platform": "ur" } ], "sha256": "bafe578518cfc46657d4ec0e74c7fe67850602ee87e075c525b460d3a2be56bc" }, "flexiv__task_0019_insert_board.zarr.zip": { "bytes": 88135246, "conversion_profile": "oxt_organizer_20261024_v1", "episodes": [ { "episode_index": 5059, "frames": 300, "platform": "flexiv" } ], "sha256": "96aa51e4d0a09f49a574350e2d860cddbdca4f50b6ebb718c91d2a3b85c15139" }, "umi__task_0133_stack_the_trays__episode_67419.zarr.zip": { "bytes": 444146275, "conversion_profile": "oxt_organizer_20261024_v1", "episodes": [ { "episode_index": 67419, "frames": 1501, "platform": "umi" } ], "sha256": "23a6a45d17201938dbc467dcc7bbf4fc80e0cf0e31677db5389529fef960858c" }, "umi_single__task_0023_cover_the_cup_with_the_lid__episode_7350.zarr.zip": { "bytes": 126175992, "conversion_profile": "oxt_organizer_20261024_v1", "episodes": [ { "episode_index": 7350, "frames": 805, "platform": "umi_single" } ], "sha256": "f4f80e021a695de4b25ebb15f8aa24b08ef69dd9a76f9fd291ef72b5d242a3cc" }, "ur__task_0047_pull_cups.zarr.zip": { "bytes": 62341290, "conversion_profile": "oxt_organizer_20261024_v1", "episodes": [ { "episode_index": 16747, "frames": 301, "platform": "ur" } ], "sha256": "eead5273ac4b9e75a009f84969bafa682a926e592d1f72b3979ca33a54d1de2b" } }
Downloads last month
237
Total file size:
2.36 GB