Dataset Viewer
Auto-converted to Parquet Duplicate
checked_at
string
frames
int64
episodes
int64
removed_episode_ids_one_based
list
removed_frames
int64
all_retained_low_dimensional_values_equal_source
bool
low_dimensional_arrays
list
sampled_image_comparisons
list
episode_boundaries_correct
bool
intra_episode_timestamp_gaps
int64
source_chunks_verified_against_published_zip
int64
view_materialization
string
encoding_runtime
dict
retained_frame_mapping_matches_independent_episode_boundary_reconstruction
bool
original_collection_episode_attribute_mapping_verified
bool
2026-11-20T11:54:07.097390+00:00
102,984
48
[ 28, 50 ]
3,060
true
[ "data/camera_main_device_timestamp_ms", "data/camera_main_frame_number", "data/camera_main_real_timestamp", "data/episode_init_joint_pos", "data/follower_joint_idx", "data/follower_joint_pos", "data/leader_joint_idx", "data/leader_joint_pos", "data/left_arm_joints", "data/left_hand_joints", "dat...
[ { "array": "data/camera_main_rgb", "frames_compared": 147 }, { "array": "data/left_tactile_data_gripper", "frames_compared": 147 }, { "array": "data/left_wrist_camera_rgb", "frames_compared": 147 }, { "array": "data/right_tactile_data_gripper", "frames_compared": 147 },...
true
0
212,947
Low-dimensional arrays materialized; image chunks generated by the same read-only filtered store used by the staged ZIP writer.
{ "blosc_threads": 1, "reason": "Deterministic byte-identical encoding required for resumable ZIP volumes; decoded data and codec configuration unchanged.", "repeatability_test": "Five encodings of the same rewritten image chunk produced one unique SHA256 with 1 thread, vs five unique hashes with 8 threads." }
true
true

TacRoboprompt_OXT

Teleoperated tabletop bimanual manipulation on an i2rt dual-arm platform, contributed to Open-X-Tactile (OXT). Each gripper has one Xsense Photon Tactile Sensor (stored identifier XenseOG); cameras comprise one fixed scene view and two wrist views.

v0.4 dataset

Task Episodes Frames
bottle 11 10,477
cup 20 24,973
lego 48 102,984
plug 13 19,121
pyramid 29 76,581
Total 121 234,136

The v0.3 release removed the specified static-right-tactile episodes from v0.2, embedded sub_task_instruction in every task Zarr, and used camera_main_rgb for the fixed scene camera. Its associated frame-number and timestamp arrays likewise use the camera_main_ prefix.

For v0.4, LEGO episodes 28 and 50 (one-based in v0.3) were removed in full due to timestamp gaps. This removes 3,060 frames. All arrays and episode boundaries use the same retained-frame mapping; timestamps are preserved without interpolation. See episode-removal provenance. LEGO structural and VLM checks were rerun after filtering.

One complete Zarr ZIP per task

Each task is one complete ZIP64 archive, split by bytes into ordered volumes because of its size. All language, camera, tactile, robot-state and episode-boundary arrays are in that task archive. No .instructions.zip overlay is required.

TacRoboprompt_OXT/
  bottle/bottle.zarr.zip.001 ...
  cup/cup.zarr.zip.001 ...
  lego/lego.zarr.zip.001 ...
  plug/plug.zarr.zip.001 ...
  pyramid/pyramid.zarr.zip.001 ...

Each task has a manifest at manifests/<task>/archive_manifest.json, outside the archive-only data directory, listing every volume's order, size and SHA256, plus the SHA256 of the complete ZIP. Download all and only the volumes listed in the current manifest into a fresh directory. Do not mix volumes from different releases. Unlike earlier independent ZIP chunks, these volumes must be concatenated before extraction:

# Example: run in the directory containing all cup volumes.
cat cup.zarr.zip.[0-9][0-9][0-9] > cup.zarr.zip
unzip cup.zarr.zip -d extracted_v04

This creates extracted_v04/cup.zarr/. Check volume hashes and the reconstructed ZIP hash against the manifest. Extract into a fresh directory so excluded episode chunks from an older release cannot remain. Current main contains the complete v0.4 LEGO archive and the four unchanged v0.3 task archives; prior versions remain in repository history.

Read and use the data

Runnable Python example: examples/read_zarr.py

python -m pip install 'zarr>=2.18,<4' numpy
python examples/read_zarr.py extracted_v04/cup.zarr --episode 0 --frame 0

The example opens the dataset in read-only mode, checks episode boundaries, reads one frame and its instruction, and demonstrates all five image streams and arm/gripper state fields without loading an entire episode. Optional --preview-dir previews writes PNGs and requires Pillow.

Zarr v2 layout:

  • data/camera_main_rgb: fixed scene RGB, (T,360,640,3).
  • data/{left,right}_wrist_camera_rgb: wrist RGB, (T,240,320,3).
  • data/{left,right}_tactile_data_gripper: (T,1,700,400,3) uint8 RGB, paired with area/sensor/type arrays.
  • data/{left,right}_arm_joints: (T,6); gripper joints (T,1) with slot index 28.
  • data/sub_task_instruction: per-frame task instruction, already inside each Zarr.
  • data/timestamps: recorded host epoch seconds; use these rather than assuming uninterrupted 30 Hz.
  • meta/episode_ends: cumulative exclusive frame endpoints.

See the FTP-1 processing format for the standard. OXT organizer metadata, annotations and check reports are available in metadata/. The final precheck reports document remaining quality warnings; they do not certify task success.

Downloads last month
477