The government plans to localise physical AI full stack technology and expand applications to manufacturing, agriculture, defence and caregiving. It sees the next 3 years as a golden time to secure competitiveness in physical AI and aims to become a global leading country that exports related technologies by 2028.

The Ministry of Science and ICT on Tuesday released a strategy to secure core competitiveness in physical AI. It plans to build a cross-ministry data collection and use system and develop key technologies such as physical AI foundation models and world models, aiming to become No. 1 in physical AI.

Physical AI goes beyond generative AI that understands language and images. It recognises real environments, predicts and reasons, and takes direct action in the real world. Robots, self-driving cars, drones and intelligent manufacturing equipment are among key applications.

The government sees physical AI as a core technology to help address social issues such as stagnant productivity, a declining population, disaster and security crises, and shortages of caregiving workers. It described the next 3 years as a golden time to secure leadership in physical AI and will focus on developing leading technologies and expanding the ecosystem, led by the science ministry.

◆ Bringing data from ministries into one place... securing proprietary physical AI models

The science ministry will first build a pan-government management system to bring together physical AI data scattered across ministries and industries. It will aggregate robot behaviour data generated through government projects and field data from manufacturing, mobility and agriculture. It will manage data quality through validity verification and interoperability standards.

It will also create an environment where companies can secure general-purpose behaviour data needed to train and test AI models. For data that are hard to obtain at real sites or involve dangerous situations, it plans to use synthetic data generated through simulations. Song Chang-jong (송창종), head of the ministry's Device AX Innovation Team, said, "It is important to make synthetic data well." He added, "In particular, synthetic data through simulators must have a configuration system that fits real sites."

It is also starting development of physical AI foundation models and world models. A physical AI foundation model is a general-purpose model that enables robots and devices to draw up task plans on their own and carry out long-duration work and precise manipulation. A world model supports AI learning and decision-making by predicting and simulating changes in the real world in advance.

A consortium including LG Electronics, Maum AI, KT, KAIST and Seoul National University has been working to secure basic technologies centred on world models since this year. The Institute for Information & Communications Technology Planning & Evaluation (IITP) will run a project through next year to develop world foundation model technology for training physical AI models, with the aim of advancing it to a level that can be used in mobile humanoids.

The government will also develop specialised devices for each field. It will develop specialised devices such as robots for poultry farming and water drones, as well as core humanoid technologies capable of fine movements at a human level. It will also develop low-latency, high-reliability networks needed for communication between robots and devices, along with end-to-end security technologies for physical AI systems. It will develop a computing platform based on ultra-low-latency, low-power, high-performance AI semiconductors so AI models can run on-device without delays.

◆ Demonstrations at manufacturing sites in South Gyeongsang and North Jeolla... expanding to defence, caregiving and agriculture

The science ministry will expand technologies proven in manufacturing to public sectors such as disaster and safety, defence and caregiving, as well as industrial sites including agriculture and mobility.

In disaster sites, it aims to apply physical AI that performs rescue and search missions in places that are difficult for people to access. In defence, it will demonstrate technologies that replace people in dangerous missions such as logistics and supply, reconnaissance, and detection and removal of weapons of mass destruction and hazardous materials.

In caregiving, it will develop robots that physically interact with people to directly perform or assist caregiving tasks. In agriculture, it will 추진 a project to automate open-field farming without humans using robots and drones.

In mobility, it will support commercialisation of self-driving cars, drones, urban air mobility and mobile robots. In manufacturing, it will build intelligent factory models using robots for small and medium-sized companies and expand them by industry, region, process and supply chain units.

The science ministry, which said it confirmed last year the feasibility of building advanced factories based on a physical AI full stack in North Jeolla and South Gyeongsang, plans to localise not only models and solutions but also the full stack including communications networks, system integration, data centres, security and operations through the second phase of the Physical AI Alliance launched last month. It plans to raise productivity in key industries by at least 20 percent.

Its goal is to become a global leading country that exports physical AI technologies in 2028. Lee Do-gyu (이도규), head of the ministry's Office of ICT Policy, said, "The goal is to maximise our technological capabilities and achieve No. 1 in physical AI through research and development and government policies."

◆ Support for GPUs, funds and talent... creating standards and testing and certification systems

The government will also push legal and institutional support, investment and talent programmes to underpin the physical AI industry ecosystem. It plans to 마련 a legal basis to comprehensively support everything from technology development to demonstrations, commercialisation and industry development. It also plans to 집중 support graphics processing units needed for physical AI training and simulation.

It will also expand investment in startups and companies' physical AI transformation projects and in long-term, challenging research and development by using the National Growth Fund, the science ministry's policy fund and the Fund of Funds. For talent development, it will pursue a 단계별 approach, from practical personnel trained through short courses and job transitions to doctoral-level advanced talent capable of developing physical AI models.

It will also establish international standards and testing and certification systems to assess the safety and reliability of technologies. It will develop benchmarks that can objectively compare the performance of robots and physical AI. It will also build safety technologies and an institutional foundation to respond to risky situations.

Bae Kyung-hoon (배경훈), deputy prime minister and minister of science and ICT, said, "It is important to secure our own technological capabilities without being dependent on the technology of global leading countries." He added, "We will actively support building a physical AI full stack based on domestic technologies and exporting it to the world after domestic demonstrations."