Beijing Humanoid Robot Innovation Center Open Sources RoboMIND Dataset with Over 50 Million Global Downloads

CGBN News Beijing, October 27th (Reporter Pang Ting) On October 26th, reporters learned from the Beijing Humanoid Robot Innovation Center that its open-source dataset RoboMIND has exceeded 50,000,000 global downloads. Since its open-source release in December 2025, the dataset has become one of the most downloaded open-source datasets in the field of embodied intelligence. Among these, the last 10,000,000 downloads were completed in just 18 days.

In terms of download growth rate, RoboMIND's first 10,000,000 downloads took 223 days, the second 10,000,000 downloads took 39 days, and the fifth 10,000,000 downloads took only 18 days, with download growth rate continuously accelerating.

Data collectors are training robots to collect data (Photo by CGBN Reporter Pang Ting)

According to the relevant responsible person from the Beijing Humanoid Robot Innovation Center, RoboMIND is the world's first open-source dataset that supports dual-arm coordination, mobile operation, dexterous hands, and high-fidelity tactile perception simultaneously. Developers can train robots to complete the full action chain from grasping, carrying to fine operations within the same dataset. This dataset is also the largest unified collection of multi-body robot operation open dataset globally, with more than 400,000 dual-arm operation trajectories released, covering more than 700 practical operation tasks.

The growth in downloads is related to the expansion of the embodied intelligence industry scale and the increasing demand for high-quality data from developers. The quality of the dataset directly affects the effectiveness of embodied intelligence applications. The more users, the more feedback, and the dataset quality continues to be optimized.

Data collectors are collecting data in a home scenario (Photo by CGBN Reporter Pang Ting)

In terms of data supply, the Beijing Humanoid Data Base covers nearly 6,000 square meters, with over 30 typical scenarios in fields such as home, supermarkets, industry, and medicine. It is equipped with 40 different configurations of robots, over 180 mainstream robot devices, and over 100 non-body data collection devices. The base uses a fully self-researched data closed-loop toolchain to connect the entire process from data collection, cleaning, quality inspection, annotation, model training, real machine evaluation, deployment, and feedback. The annual data production capacity reaches 180,000 hours, with nearly 30,000 hours of high-quality data already delivered. The Beijing Humanoid Robot Innovation Center has also led the development of the country's first industry standard for embodied intelligence datasets.

In addition to open-source datasets, the Beijing Humanoid Robotics Innovation Center has also launched the “Huishi Kaiwu” application store. Users do not need to start from scratch; they can download skills and deploy them to robots with a single click. Developers can upload videos to automatically generate skills, or upload their self-developed skills for sharing.

A relevant official from the Beijing Humanoid Robot Innovation Center stated that looking to the future, the center is willing to collaborate with more partners to build, share, and jointly research data, focusing on areas such as industry data collaboration, scenario data co-creation, and joint model training, to promote the development of embodied intelligence. Return to the previous page for more information