The World’s First End-to-End Embodied AI Data and Training Platform

Beijing, China - [Dec, 2025] - RealMan Robotics, a pioneer in ultra-lightweight humanoid robotic arms, recently announced the open-source release of RealSource, the world's first high-quality, largest-modality multi-modal real-world robot dataset, marking a major milestone in breaking data barriers in embodied intelligence. Designed to address the industry's shortage of fully aligned real-world data, the dataset provides essential support for next-generation robot perception, planning, and control algorithms in both academia and industry.

The dataset is built entirely on 10 real-world simulated environments within the Beijing Humanoid Robot Data Training Center and features two defining characteristics: exceptional data quality and the world's most complete multi-modal coverage.

Project homepage: https://realmanrobot.github.io/real_source_dataset
Open-source access: https://huggingface.co/datasets/RealSourceData/RealSource-World

High-Fidelity Data Across 10 Real-World Scenarios
The dataset originates from the 3,000 m² Beijing Humanoid Robot Data Training Center, featuring:

Training Zone: High-volume, efficient robot training for foundational manipulation tasks.
Scenario Zone ("Robot University"): Ten real-world environments including smart home & eldercare, daily living, agriculture, new retail, automotive assembly, and catering.

Robots perform lifelike tasks such as opening refrigerator doors, folding laundry, and sorting materials on factory lines, capturing data in realistic, noisy, and diverse environments.

Data collection is conducted outside the "laboratory greenhouse", directly addressing the complexity, noise, and diversity of real-world environments. This ensures high realism, strong practicality, and superior generalization across scenarios.

Key metrics: 100% modality completeness, 78% noise-resistance, 82.1% smoothness, setting a new benchmark for real-world embodied intelligence datasets.

Multi-Modal Data with Six Core Advantages
The dataset covers the full perception-decision-execution chain, integrating RGB images, joint angles & velocities, six-axis force, end-effector pose, action commands, timestamps, and camera parameters.
Hardware-Level Spatiotemporal Synchronization: All sensors aligned to a unified physical coordinate system.
Ultra-Low Frame Loss (<0.5%) : Continuous, reliable recording even at high speed.
High-Precision Motion Control: Millisecond-level joint data for smooth, accurate operations.
Factory-Calibrated for Out-of-Box Use: No extra calibration required.
Generalization-Oriented Collection: Tasks repeated under diverse object, environment, and lighting conditions.
Exoskeleton Teleoperation: 1:1 human-to-robot motion mapping for high-fidelity demonstration.
High-Performance Data Collection Platforms
RealMan's robots are designed to match adult human arm proportions, achieving seamless integration with real-world tasks:

Payload: 5 kg rated, up to 9 kg max
TCP speed: 1.8 m/s
Power consumption: <100 W
MTBF: 50,000 hours

Robots used for data collection:
RS-01: Wheeled Folding Mobile Robot, 20-DOF dexterity, multi-modal vision.
RS-02: Dual-Arm Lifting Embodied Robot, RGB + depth vision, dual 7-DOF arms, 9 kg payload per arm, Six-axis force sensing + overhead fisheye perception.
RS-03: Dual-Arm Dual-Eye Precision Robot, upgraded binocular system for high-resolution stereo vision and precise manipulation.

All three robots integrate large FOV wrist and head cameras (H 90° / V 65°) and full spatiotemporal synchronization, forming a high-performance collection platform suitable for industrial automation, home service robotics, and scientific research.

Toward a Global Open-Source Robotics Ecosystem
By releasing RealSource, RealMan Robotics aims to break data silos and accelerate global embodied intelligence research. The company plans to continue expanding the dataset, adding scenarios and modalities, and building a fully open, interconnected ecosystem that bridges research and industrial deployment.