Motion data and world model technologies to support autonomous humanoids capable of complex real-world tasks
NC AI is expanding its physical AI business into humanoid robotics, joining a consortium with LG Electronics to develop autonomous robots capable of learning human movements and performing complex tasks in real-world environments.
The physical AI company said on Sept. 3 that the consortium has been selected to participate in the humanoid segment of Korea’s “K-On-Device AI Semiconductor Technology Development Project,” led by the Ministry of Trade, Industry and Energy and the Korea Evaluation Institute of Industrial Technology (KEIT).
The project aims to develop autonomous humanoid robots that can perform human-level tasks in complex everyday environments using domestically developed on-device AI semiconductors.
NC AI will be responsible for building initial training data for learning-based General Motion Tracking (GMT) and developing a Human-to-Robot Retargeting pipeline that converts human movements into motions a humanoid robot can physically perform.
Turning human movement into robot training data
Transferring human motion to a humanoid involves more than simply copying movements. Humans and robots have different body proportions, joint structures and ranges of motion.
NC AI’s retargeting technology will therefore take into account physical constraints such as joint limits, body proportions, foot contact and balance. Human work movements can then be converted into training data optimized for a humanoid’s physical structure and used for whole-body control.
Such technology is becoming increasingly important as the robotics industry shifts from machines designed to repeat predetermined actions toward physical AI systems that perceive their surroundings, make decisions and act autonomously.
For humanoid robots in particular, advances in hardware alone are not enough. AI models and large-scale training data must ultimately be translated into stable physical movements in unpredictable real-world environments.
Motion technology expands from digital content to physical AI
NC AI plans to apply its experience in motion data and animation production to the robotics field.
The company has accumulated technologies for capturing and digitizing complex human movements through years of working with large-scale motion data and advanced animation pipelines. It now plans to use that expertise to convert a wider range of human activities into datasets suitable for humanoid training.
NC AI will also create a digital twin environment to generate large-scale motion datasets across diverse scenarios without relying entirely on physical robot training.
The approach could allow humanoids to encounter far more situations during training while reducing the time and physical constraints involved in repeatedly collecting data with real robots.
The company plans to combine these motion-control capabilities with its world model technology. The longer-term goal is to enable robots not only to reproduce learned movements but also to anticipate changes in their surroundings and autonomously execute complex, multi-step tasks.
From household tasks to manufacturing and disaster response
The consortium brings together robot hardware, AI models, training data and domestically developed on-device AI semiconductors within a single development framework.
On-device processing is particularly relevant for autonomous robots because AI computations can be performed directly on the robot rather than depending entirely on external servers. The project aims to secure high-performance, low-power AI semiconductor technology capable of supporting real-time decision-making and motion control.
The resulting technologies are expected to be applied first to general-purpose humanoids performing tasks such as organizing and transporting objects, serving items and handling tools in unstructured environments including homes.
Potential applications could eventually extend to manufacturing and logistics support, facility inspection, disaster response and work in hazardous locations where autonomous decision-making and sophisticated physical control are required.
NC AI is also expanding its physical AI business around world models and robot foundation models, with the humanoid project providing another route to connect its AI technologies with machines operating in the physical world.
“The significance of this project lies in bringing the motion-control technology and data-building capabilities that NC AI has accumulated over many years into the autonomous movement of actual humanoid robots,” NC AI CEO Lee Yeon-soo said.
“We will work closely with LG Electronics and other partners to maximize the competitiveness of humanoids powered by domestically developed on-device AI semiconductors and contribute to securing Korea’s strategic technology capabilities,” Lee added.
