South Korea opened its first dedicated data-collection facility for humanoid robots on November 3, a joint project between the Korea Institute of Machinery and Materials (KIMM) and the Electronics and Telecommunications Research Institute (ETRI). The 'AI Humanoid Data Factory' currently runs 12 humanoid robots and is built to generate the real-world task data Korean developers need to train robot AI models without depending on datasets produced overseas.
What does the facility actually do?
The Data Factory collects data as robots perform everyday and industrial tasks, including picking up objects, handling laundry, retrieving items from refrigerators, rearranging furniture, and operating vehicle buttons and levers. Each action is logged and used to train AI models intended to let humanoid robots act autonomously in unstructured environments rather than follow pre-programmed routines.
At the opening ceremony, the Korean-developed humanoid KAIROS demonstrated precise object manipulation and full-body motion imitation, giving officials a public showcase of the kind of behavior the facility's data is meant to produce at scale.
Why is Korea building this now?
The facility operates under the Global TOP Strategic Research Group for Self-Growing AI Humanoids, led by KIMM researcher Chan Hun Park, and is funded through the Ministry of Science and ICT's K-Moonshot AI Humanoid Mission. KIMM President Seog-Hyeon Ryu said 'securing large volumes of high-quality robot task data is essential to national technological competitiveness,' calling the opening 'an important starting point for building Korea's own robot AI training data and technological assets.'
The motivation is straightforward: the humanoid robots now being built by US and Chinese developers mostly learn from data collected on those companies' own hardware and in their own facilities. Korea's robotics industry, strong in manufacturing and components but smaller in AI, risks ending up a hardware supplier with no say over the datasets that will determine which robots become competent enough to work unsupervised. A state-run data pipeline reduces that dependency.
How does this compare to what other countries are doing?
The approach mirrors moves by China's robotics sector, where firms have poured billions of dollars into AI 'brains' for humanoids this year specifically to build proprietary training datasets, and by companies like Nvidia, which has built simulation tools to generate synthetic training data at scale. Korea's answer is more centralized: a single government-backed facility rather than dozens of companies each building isolated datasets, with KIMM positioning itself as an infrastructure provider that the entire domestic industry can eventually draw from.
That centralization is also the facility's biggest limitation for now. With only 12 robots in operation, the Data Factory is tiny next to the fleets Chinese manufacturers already have collecting data across factories, stores and warehouses. KIMM has not given a target robot count, saying only that it plans 'gradual expansion.'
What happens next?
KIMM says the facility will remain restricted to its own research group before opening to domestic industry, academia and other research institutions starting at the end of 2027. That delay gives Korean companies working on humanoids, including Rainbow Robotics and the broader K-Humanoid Alliance, a multi-year runway before they can tap the resulting datasets directly. Whether that timeline is fast enough to matter depends on how quickly rival training-data efforts in China and the US scale over the same period.