en.Wedoany.com Reported - China's RealMan Robotics has released the open-source dataset RealSource, designed to provide high-quality, multimodal training data for AI-driven robot control research and break down data silos within the industry. The dataset is built upon ten real-world application scenarios from the Beijing Humanoid Robot Data Training Center, emphasizing high data fidelity and cross-scenario generalization capabilities.
The dataset originates from two areas of the 3,000-square-meter training center. In the training area, robots perform basic manipulation tasks; the scenario area includes ten real-world application environments such as smart home and elderly care, daily applications, agriculture, retail, catering, and automotive production, where robots execute tasks like opening refrigerator doors, folding clothes, and sorting materials. The high-fidelity, noisy, and diverse environments remove data collection from laboratory conditions, with key metrics including 100% modality completeness, 78% noise resistance, and 82.1% smoothness.
The dataset covers the entire process of perception, decision-making, and execution. Collected data includes RGB images, joint angles and velocities, six-axis forces, end-effector positions, action commands, timestamps, and camera parameters. Its features include hardware-based spatiotemporal synchronization, extremely low frame loss rate (less than 0.5%), high-precision millisecond-level joint motion control, factory calibration for immediate use, generalizable task collection across different objects and environments, and 1:1 human-to-robot motion mapping via exoskeleton.
Three robot types were used for data collection. The RS-01 is a mobile robot with wheels and a folding mechanism, featuring a 20-degree-of-freedom arm and multimodal vision. The RS-02 is a dual-arm robot suitable for handling tasks, with vision including RGB and depth perception, each arm having 7 degrees of freedom and a 9 kg payload, and collection utilizing six-axis force sensors and an overhead fisheye camera. The RS-03 is a dual-arm precision robot equipped with a binocular system for high-resolution stereo vision. All robots are equipped with wrist and head cameras, with a 90-degree horizontal field of view and a 65-degree vertical field of view, and feature full spatiotemporal synchronization.
Realman's robotic arms are proportioned to match adult human arms, with a rated payload of 5 kg, a maximum payload of 9 kg, a TCP speed of 1.8 m/s, power consumption below 100 watts, and a Mean Time Between Failures (MTBF) of 50,000 hours. RealMan Robotics plans to bridge research and industrial applications by open-sourcing the RealSource dataset, and will further expand the dataset in the future by adding new scenarios and modalities to build an open, interconnected ecosystem.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
