Realman Introduces Realsource Multimodal Dataset for Robotics Applications

Related Vendors

FAULHABER Drive Systems

Taiwan Machine Tool & Accessory Builders' Association (TMBA)

SIMTOS (Seoul International Machine Tool Show)

Realsource was developed by robot manufacturer Realman as an end-to-end platform for robot data and training with integrated AI, and is available as open source.

The dataset was collected and compiled because the industry lacks comprehensive and consistent data of this kind. It provides important support for the development of the next generation of algorithms for controlling robots in science and industry.

The dataset is based exclusively on ten simulations of real-world application environments at Realman's Beijing Humanoid Robot Data Training Center. It is characterized by exceptionally high data quality and the world's most comprehensive multimodal coverage.

Gallery

High-Precision Data from Ten Realistic Applications

The dataset comes from two areas of the 32,000 square feet Beijing Humanoid Robot Data Training Center. The training zone is used for extensive robot training in basic manipulation tasks. The scenario zone (“Robot University”) features ten real-world environments for robotics applications, including smart homes and elder care, everyday applications, agriculture, retail, food service, and automotive manufacturing. Here, robots perform realistic tasks from both household and industrial settings, such as opening refrigerator doors, folding laundry, and sorting materials on production lines.

This work takes place in realistic, noisy, and diverse environments. As a result, data collection occurs outside the “laboratory greenhouse.” This yields highly practical datasets with excellent generalizability across various scenarios. Key metrics include: 100% modality completeness, 78% noise resistance, and 82.1% smoothing.

Six Benefits of Multimodal Data

The dataset covers the entire process chain from data acquisition to decision-making and execution. Among other things, it includes RGB images, joint angles and velocities, six-axis forces, end-effector positions, action commands, timestamps, and camera parameters. Here are some specific features in detail:

• Hardware-based spatiotemporal synchronization: All sensors are aligned to a uniform physical coordinate system.
• Extremely low frame loss (<0.5%): Continuous, reliable recording even at high speeds.
• High-precision motion control: Joint data updated in milliseconds for smooth, accurate operations.
• Factory-calibrated for immediate use; no additional calibration required.
• Generalizable assessment: Tasks are repeated under various object, environmental, and lighting conditions.
• Exoskeleton remote control: 1:1 motion mapping from human to robot enables highly precise reproduction.

Realman's robotic arms are proportioned like those of an adult human and can be seamlessly integrated into real-world tasks. They have a rated load capacity of 11 pounds and can lift up to 20 pounds. The TCP speed is 6 feet per second, with a power consumption of less than 100 watts and a mean time between failures (MTBF) of 50,000 hours.

Three types of robots were used for data collection. The RS-01 is a mobile robot with wheels and a folding mechanism, an arm with 20 degrees of freedom, and multimodal vision. The RS-02 is a two-armed robot designed for lifting tasks; its vision system includes RGB plus depth perception. Each of the two arms has seven degrees of freedom and a payload capacity of 20 pounds. Data acquisition is performed using a six-axis force sensor system and overhead fisheye imaging. The RS-03 is a two-armed precision robot with a high-performance binocular system for high-resolution stereo vision and precise manipulation.

All three robots are equipped with wrist- and head-mounted cameras featuring a wide field of view—90 degrees horizontally and 65 degrees vertically—as well as full spatiotemporal synchronization. They represent a powerful robotics platform suitable for industrial automation, household applications, and scientific research. The datasets generated by them are correspondingly diverse.

By releasing Realsource as an open-source dataset, Realman Robotics aims to break down data silos and advance research in the field of AI-powered robot control. The company plans to further expand the dataset, add new scenarios and modes, and build a fully open, interconnected ecosystem that bridges the gap between research and industrial applications.

Design simulation for industrial robots

How to make your production more flexible with robotics