Shanghai Robotics Firm Releases Open-Source Humanoid Robot Dataset

Shanghai-based robotics firm Fourier Intelligence has released an open-source full-size humanoid robot dataset called Fourier ActionNet, along with what it describes as the world's first full-process tool-chain, according to an announcement on the company's WeChat account Monday.

The initiative aims to enhance AI robot training and provide a comprehensive solution for developers and research institutions worldwide, fostering innovation and collaboration within the global robotics community.

The initial dataset includes over 30,000 high-quality real-machine training entries covering dexterous hand movements and specialized imitation learning data for hand-related tasks. These applications span scenarios such as picking up and laying down tools, performing household tasks, and other common actions.

"High-quality robot motion data is essential for advancing embodied intelligence," noted a robotics expert quoted in the announcement. Open-sourcing this data allows researchers and developers to share insights and methodologies, potentially improving robot adaptability in dynamic environments and strengthening human-robot interaction.

Jiang Lei, chief scientist of the Shanghai-based National and Local Co-Built Humanoid Robotics Innovation Center, told the Global Times that collecting real-world robot motion data has been constrained by high costs and low annotation accuracy, which has limited industry progress. Jiang emphasized that fostering an open-source culture in embodied intelligence can spur innovation and drive academic breakthroughs.

Fourier Intelligence has collaborated with over 20 top domestic and international research institutions and leading industry enterprises. The company plans to release more advanced data modules covering full-body motion control and multi-task coordination in the future.

Other Chinese robotics startups are making similar moves. Shanghai-based AgiBot has launched its dataset named AgiBot Digital World, comprising more than 1 million robot action trajectories captured from 100 robots in real-world scenarios. Additionally, Hangzhou-based Unitree has updated its H1 and G1 humanoid robots with an open-source full-body dataset, using motion capture data to make movements appear more natural and smoother.

Reported by Zhang Weilan, Global Times