Skip to content

Dataset · RealmanRobot

Dual-arm humanoid robot manipulation dataset

RealmanRobot/RealSource-World

Over 14 million frames and 11,400 RS-02 robot control episodes for real-world tasks.

What's inside

A large-scale multimodal robotics dataset recorded in real environments using an RS-02 dual-arm humanoid robot. Contains over 14 million video frames and more than 11,428 episodes across 35 different manipulation tasks. Data features 71-dimensional parameters, atomic skill annotations, quality scores, and English and Chinese text.

What you can do with it

  • Training Embodied AI models and robot control systems
  • Training Vision-Language-Action (VLA) models for complex manipulations
  • Decomposing long action sequences into atomic skills
Tasks
Robotics
Languages
English, Chinese
Size
10K–100K rows
Downloads this month
89.9K
Likes
32
Access
on request on Hugging Face

How to load it

from datasets import load_dataset

ds = load_dataset("RealmanRobot/RealSource-World")