Joshua69/vt-muse-oxt

Dataset

VT-MUSE Real-World Visuotactile Manipulation Dataset

0

2 commits

2 linked in READMEs

updated Aug 21, 2026

See the code

README

VT-MUSE Real-World Visuotactile Manipulation Dataset

This repository hosts the real-world tactile manipulation data contributed with VT-MUSE to the Open-X-Tactile (OXT) dataset.

The release follows the FTP-1-compatible Zarr organization required by OXT. Each task is distributed as an independently downloadable ZIP archive containing one standardized Zarr dataset.

Dataset contents

Sub-datasetTrajectories
insert_tube250
press_toaster100
pull_out_drawer100
wipe_board201
Total651

The standardized release contains 192,225 temporally aligned frames. It includes the 200 trajectories used for the final VT-MUSE real-world experiments (50 per task) and 451 additional valid real-world trajectories. The authoritative episode-level split is recorded in the OXT submission manifest.

Repository layout

insert_tube/
  insert_tube.zarr.zip
press_toaster/
  press_toaster.zarr.zip
pull_out_drawer/
  pull_out_drawer.zarr.zip
wipe_board/
  wipe_board.zarr.zip
README.md
supplementary_tactile_note.json

Sensors and observations

  • Fixed third-person RGB camera, exported as camera_main_rgb.
  • Robot state/action: seven arm joints and one gripper degree of freedom.
  • Two GelSight Mini tactile pads, with aligned tactile RGB observations.
  • Supplemental per-pad tactile depth maps are documented in supplementary_tactile_note.json.

Format and quality control

The archives use the OXT/FTP-1-compatible Zarr schema. Metadata and structure validation was run with the official OXT-QC toolkit. The four task-level Zarr datasets passed the metadata precheck with no reported errors or warnings.

For the data-format specification and the broader collection, visit the Open-X-Tactile website.

Institution and contact

Shanghai Jiao Tong University
Contact: acondaway@sjtu.edu.cn

manipulation
open-x-tactile
robotics
tactile-sensing
visuotactile

Joshua69/vt-muse-oxt

Dataset

VT-MUSE Real-World Visuotactile Manipulation Dataset

0

2 commits

2 linked in READMEs

updated Aug 21, 2026

See the code

README

VT-MUSE Real-World Visuotactile Manipulation Dataset

This repository hosts the real-world tactile manipulation data contributed with VT-MUSE to the Open-X-Tactile (OXT) dataset.

The release follows the FTP-1-compatible Zarr organization required by OXT. Each task is distributed as an independently downloadable ZIP archive containing one standardized Zarr dataset.

Dataset contents

Sub-datasetTrajectories
insert_tube250
press_toaster100
pull_out_drawer100
wipe_board201
Total651

The standardized release contains 192,225 temporally aligned frames. It includes the 200 trajectories used for the final VT-MUSE real-world experiments (50 per task) and 451 additional valid real-world trajectories. The authoritative episode-level split is recorded in the OXT submission manifest.

Repository layout

insert_tube/
  insert_tube.zarr.zip
press_toaster/
  press_toaster.zarr.zip
pull_out_drawer/
  pull_out_drawer.zarr.zip
wipe_board/
  wipe_board.zarr.zip
README.md
supplementary_tactile_note.json

Sensors and observations

  • Fixed third-person RGB camera, exported as camera_main_rgb.
  • Robot state/action: seven arm joints and one gripper degree of freedom.
  • Two GelSight Mini tactile pads, with aligned tactile RGB observations.
  • Supplemental per-pad tactile depth maps are documented in supplementary_tactile_note.json.

Format and quality control

The archives use the OXT/FTP-1-compatible Zarr schema. Metadata and structure validation was run with the official OXT-QC toolkit. The four task-level Zarr datasets passed the metadata precheck with no reported errors or warnings.

For the data-format specification and the broader collection, visit the Open-X-Tactile website.

Institution and contact

Shanghai Jiao Tong University
Contact: acondaway@sjtu.edu.cn

manipulation
open-x-tactile
robotics
tactile-sensing
visuotactile