Skip to content

Dataset · HuggingFaceVLA

Cross-platform robot dataset from Hugging Face

HuggingFaceVLA/community_dataset_v3

Large open video dataset for training Vision-Language-Action models across 46 robot types.

What's inside

Massive robotics dataset collected by 235 community participants worldwide based on datasets for SmolVLA. Includes 791 subsets (from 10k to 100k records) with video recordings and English text instructions. Data covers 46 different robot types, enabling universal control policy training.

What you can do with it

  • Pretraining Vision-Language-Action (VLA) models for robot control
  • Developing universal algorithms transferring skills between different robot types
  • Fine-tuning robotic systems to execute commands via video and text
Tasks
Robotics
Languages
English
Size
10K–100K rows
Downloads this month
81K
Likes
25

How to load it

from datasets import load_dataset

ds = load_dataset("HuggingFaceVLA/community_dataset_v3")