Dataset

TouchScale

Contains 100 hours of egocentric human manipulation with synchronized RGB-D, dual wrist video, and full-hand tactile signals for visuo-tactile learning.

Hugging Face ↗Sep 30, 2026Paper ↗

  • robotics
  • visuo-tactile
  • egocentric
  • tactile
Overview figure showing data collection, dataset scale, and evaluation tasks of TouchScale
Overview figure showing data collection, dataset scale, and evaluation tasks of TouchScale

What it is

Large egocentric visuo-tactile manipulation dataset with synchronized RGB-D, wrist video, and dense tactile glove signals from human demonstrations. 100 released hours of egocentric human manipulation data, scaling to 500 hours, with a unified rig: head RGB-D, two wrist cameras, and full-hand tactile gloves with 880 taxels.

Key results

  • 100 hours released; 500 hours planned
  • 1 head-mounted RGB-D camera + 2 wrist cameras
  • Full-hand tactile gloves with 880 taxels

Useful for training and evaluating visual-tactile models for everyday manipulation from human demonstrations.

One useful ML email a week.

Worth your time, not all of your time.

We track new ML work across papers, code, and data, then keep only the pieces worth a closer look. Selected, not collected.

Papers.Data.Code

The Weekly ML Signal

Selected, not collected.

One useful ML email a week: the papers, repos and datasets worth your attention.

Worth your time, not all of your time.