South Korea’s new humanoid robot data factory teaches real-world tasks

For the first time ever, the South Korean government launched a humanoid robot data factory that leverages 12 bipedal robots to gather physical motion and operational data across household and industrial settings.

​The Korea Institute of Machinery and Materials (KIMM) unveiled the facility to support the development of autonomous physical AI systems in collaboration with the Electronics and Telecommunications Research Institute (ETRI) on November 3.​

As the first such data generation laboratory in the country, the facility gathers high-resolution movement data across robotic arms, legs, and full-body systems. Human operators remotely guide the machines to undertake tasks of varying degrees of complexity, recording spatial coordinates, mechanical torque, and tactile feedback for each action.​

The rich empirical dataset of recorded operational routines will be used to train large AI models to better comprehend advanced directives from humans and understand surrounding environments, with the eventual goal of allowing humanoid robots to execute these tasks autonomously.

Demonstrating physical AI in domestic and industrial scenes

At the demonstration events for the inauguration, the participating humanoids conducted several household responsibilities such as opening refrigerator doors to retrieve items for users, arranging living room furniture, and handling laundry.

​In industrial demonstrations, the robots showed off operational movements inside a vehicle, manipulating levers and control buttons to engage the systems.​ The event also featured KAIROS, an advanced, domestically developed humanoid platform that replicated sophisticated object manipulation and real-time full-body motion imitation.

KIMM stated that neural network models are tested on physical machines, and telemetry from these tests is continuously fed back into the training pipeline to refine future iterations. As an overarching technical goal, the institute aims to leverage the data to build generalized AI architectures applicable across diverse robotic chassis, including the country’s K-Humanoid standard platform.

Fostering data sovereignty and ecosystem access in South Korea

Led by KIMM principal researcher Chan Hun Park, the data factory operates under the Global TOP Strategic Research Group for Self-Growing AI Humanoids. KIMM emphasized that gathering a large dataset of in-house physical training data will prevent South Korea from relying on foreign foundation models and proprietary overseas datasets.​ To advance these ambitions, the initiative is supported by the Ministry of Science and ICT as a key project under its K-Moonshot AI Humanoid Mission.

KIMM plans to expand its physical fleet beyond the existing 12 robots and continue gathering operational data across more complex environments like assembly lines and logistics hubs.​

“Securing large volumes of high-quality robot task data is essential to national technological competitiveness, and the establishment of this Data Factory marks an important starting point for building Korea’s own robot AI training data and technological assets,” said Seog-Hyeon Ryu, president of KIMM.​ Starting in late 2027, KIMM will begin to provide operational access to domestic enterprises, academia, and independent researchers, who will collaborate to strengthen national competitiveness in robot AI on the international stage.

Source: https://interestingengineering.com/ai-robotics/south-korea-humanoid-robot-data-factory