China scales humanoid robot industry with massive data collection efforts

China's humanoid robot industry is experiencing rapid hardware development, with domestic sales projected to triple to 50,000 units this year. While companies like Unitree have achieved significant mobility milestones, such as running speeds of 12.66 m/s, the industry faces challenges regarding software capabilities and high development costs.

To transition from simple movement to complex tasks, developers must expand foundation models from billions to hundreds of billions of parameters. This requires massive amounts of real-world physical data—such as spatial awareness and the ability to sense object fragility or liquid viscosity—which is much harder to collect than digital text used for AI chatbots.

To address this data shortage, China is implementing a large-scale, state-supported initiative to collect physical data through two primary methods. The first involves using specialized training centers where robots perform human-like tasks, such as shelving goods or working on factory lines. Currently, there are 53 such centers in operation, with plans to build or develop an additional 34.