Skip to content

Dataset · FVL-Lab

Visuo-tactile benchmark for robotics

FVL-Lab/OpenViTac

A dataset of 6,330 episodes for training and evaluating vision- and touch-based robotic arm models.

What's inside

The dataset contains multimodal data (tables, text, video) with a total volume of 1K to 10K episodes (6,330 in total: 1,500 simulations and 4,830 real experiments). It describes fragile object manipulation and precise assembly tasks using GelSight mini, Neote, and Xense sensors in LeRobot v2.1 format.

What you can do with it

  • Training policy models for robotic manipulations
  • Evaluating tactile perception and fragile object interaction models
Size
1K–10K rows
Downloads this month
20.2K
Likes
0

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("FVL-Lab/OpenViTac")