Dataset Viewer

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.

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

RMtac — Tactile UMI Manipulation Dataset

Handheld UMI (Universal Manipulation Interface) demonstrations with dual FlexiTac tactile fingertips, contributed to the Open-X-Tactile (OXT) collection in the FTP-1 zarr format.

Tasks

Task Episodes Frames Instruction
pick_crack_egg 104 45,062 pick up the egg, adapting the pick position, then crack the egg
pour_oil_pan 122 50,295 pick up the oil bottle, squeeze it to pour oil into the pan
scoop_egg_toast 117 78,084 pick up the spatula, scoop the fried egg from the pan and place it on the toast
total 343 169,051

Contents

Each RMtac/<task>.zarr.zip is a zarr v2 store (zip entries at the archive root, readable directly via zarr.ZipStore) with the FTP-1 data/ arrays:

  • right_wrist_pose (T, 6) — xyz + rotvec in the OXT UMI canonical gripper frame (z forward, x down, y left), first frame = identity per episode
  • right_wrist_camera_rgb (T, 224, 224, 3) — GoPro fisheye wrist view
  • right_tactile_data_gripper (T, 2, 12, 32) — raw FlexiTac counts, area 0 = left pad, 1 = right pad (right_tactile_area_gripper)
  • right_hand_joints / robot0_gripper_width_m (T, 1) — gripper width in metres, joints idx 28
  • sub_task_instruction, timestamps, meta/episode_ends

Collected at 30 Hz with a RealSense T265 (pose), GoPro 2.7K 4:3 via HDMI capture (wrist camera) and two FlexiTac 12x32 sensors at ~125 Hz, hardware-timestamped and aligned per frame.

Format reference

This dataset is a contribution to Open-X-Tactile (OXT): https://open-x-tactile.github.io/ — see the OXT site for the FTP-1 data format specification and the full collection.

Affiliation / contact

UC San Diego & San Diego State University — xixinzhang96@gmail.com

Downloads last month
33