The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.
DexManiTac Dataset
Open X-Tactile / FTP-1 compatible manipulation data collected with a Franka FR3, parallel gripper, Intel RealSense cameras, and one Xense vision-based tactile sensor.
Hugging Face: https://huggingface.co/datasets/xinpan/DexManiTac
Summary
- 5 tasks
- 337 trajectories
- 29,802 synchronized frames (approximately 49.7 minutes at 10 Hz)
- Xense installed on the left gripper pad (
functional-area ID = 0) - Contact:
mengxinpan2024@ia.ac.cn
| Task | Trajectories | Frames |
|---|---|---|
| In-hand relocation | 124 | 12,268 |
| Pull bottle | 60 | 4,010 |
| Stamp | 65 | 6,418 |
| Wipe board | 25 | 3,270 |
| Wipe board with a pre-grasped tool | 63 | 3,836 |
Contents
Each task is stored as a Zarr v2 directory with data/ arrays and
meta/episode_ends. The main fields are:
camera_main_rgb: D435 RGB, 224 x 224right_wrist_camera_rgb: D405 RGB, 224 x 224right_arm_joints: seven FR3 joint positionsright_hand_joints: parallel-gripper width, canonical index 28right_wrist_pose: xyz in the FR3 base frame plus rotation vectorright_tactile_data_xense_image: Xense RGB tactile imageright_tactile_data_xense_force_grid: 35 x 20 x 3 dense force fieldright_tactile_data_xense_force_resultant: six-dimensional force resultant- paired
right_tactile_area_*,right_tactile_sensor_*, andright_tactile_type_*metadata arrays sub_task_instructionandtimestamps
All camera and tactile images are RGB. Modalities were nearest-neighbor aligned to the host timestamp using their recorded source timestamps, then cropped to common temporal support. Source HDF5 files were not modified.
In the Hugging Face Files and versions view and in ModelScope, task stores are packaged under one dataset parent directory rather than exposed at the repository root:
DexManiTac/
|-- inhand_relocation.zarr.zip
|-- pull_bottle.zarr.zip
|-- stamp.zarr.zip
|-- wipe_board.zarr.zip
`-- wipe_board_simple_pregrasp.zarr.zip
Quality control
validation_report.json records the automated schema, episode-boundary, timing,
shape, metadata, and full floating-point finite-value checks. source_manifest.csv
provides per-episode provenance and synchronization statistics. The qc/ directory
contains visual montages, and video.mp4 is a short multi-task preview.
Conventions
The sensor identifier is Xense. Its functional area is confirmed as the left
gripper pad / thumb side (0). right_wrist_pose preserves the FR3-base world
frame and the collected end-effector local-axis convention; no additional fixed
transform to another gripper convention has been applied.
License
License to be confirmed before public release.
- Downloads last month
- 73