(Yicai) April 17 -- China’s humanoid robot unicorn AgiBot has launched a new subsidiary, Maniformer, which unveiled a one-stop data service platform and data collection hardware to address the massive data demands for training embodied intelligence models, with a target of producing tens of millions of hours of data this year.

The move highlights the growing bottleneck in high-quality data supply for embodied artificial intelligence, as the industry currently has only about 500,000 hours of such data globally. Maniformer aims to bridge this gap by building scalable infrastructure and leveraging a global data collection network to rapidly expand supply, Yicai learned at a launch ceremony in Shanghai yesterday.

Maniformer’s physical AI data service platform will provide real machine data, simulated data, and human demonstration data, enabling systematic, standardized, and scalable data production, the company said. Unlike large language models trained on vast internet data, embodied AI models rely on real-world interaction data, which remains scarce, said Yao Maoqing, chairman and chief executive of Maniformer. Yao is also a partner and senior vice president of Shanghai-based AgiBot.

To lower barriers and costs in data collection, Maniformer also introduced the MEgo series of hardware, including data collection grippers, head-mounted devices, and a data governance engine. The system supports an industry-first approach to data collection without a physical robot, where human operators wear sensor devices to perform tasks in real environments while capturing motion trajectories, visual inputs, and tactile feedback for standardized use across different robots.

At the event, Maniformer also launched a data co-creation initiative, with dozens of companies and institutions, including the Beijing Innovation Center of Humanoid Robotics, joining the effort. The initiative aims to build the world’s largest physical AI data ecosystem and achieve a data production capacity of 10 billion hours by 2030.

Editors: Dou Shicong, Emmi Laine