Physical AI is emerging as a core technology that can be applied in real-world industrial settings—from the factory floor to field operations—to boost industrial competitiveness. It is expected to play a pivotal role in advancing major national policies and the administration’s key policy agenda over the next four to five years. On Thursday, August 14, the Korea Institute of S&T Evaluation and Planning (KISTEP), led by President Tae-seog Oh, hosted the National Strategic Technology Seminar on Physical AI at its headquarters in Chungbuk Innovation City.

The seminar was organized to explain the concept of Physical AI, assess its potential applications in industry and explore strategic directions to strengthen Korea’s industrial competitiveness. Byung-hee Son, director of the Maeum AI Research Institute, delivered an in-depth presentation on the latest in Physical AI, including its social backdrop and necessity, real-world use cases and global trends.

△ Tae-seog Oh, President of KISTEP, delivering opening remarks

Son defined Physical AI as “AI that operates in the real world where the laws of physics apply” and discussed its potential to expand into a wide range of sectors such as robotics, autonomous vehicles (AVs), agriculture, defense and public services. Highlighting case studies—including an unmanned pesticide sprayer, an autonomous air purifier and an accessible kiosk—he argued that Physical AI can help address labor shortages driven by population aging and decline. He went on to identify on-device AI, model compression and optimization, and physics-informed learning as core competitive strengths. Finally, he called for developing domain-specific models using digital twins and simulation, and for targeting global niche markets through partnerships with small and medium-sized enterprises (SMEs).

△ Byung-hee Son, Director of the Maeum AI Research Institute

Seminar Highlights

1. What Physical AI Is and Why It Matters

Unlike traditional AI, which is confined to digital environments such as text, images and video, Physical AI applies the laws of physics to operate in the real world. It goes beyond simple simulation or video generation, enabling robots and machines to perform tasks and take action in real-world settings. Generative AI focuses on content creation, AI agents on decision-making and task execution in digital environments, and Physical AI on robots and automation systems operating in the physical world. The need for physical AI is growing, especially to address labor shortages amid population decline and to ensure safety and efficiency across industries, including manufacturing, agriculture and defense.

2. Core Technical Pillars 

○ On-device AI: Runs AI models locally on the device without relying on a network connection, ensuring continued operation in unstable environments while strengthening security and privacy. A notable example is the world’s first attempt to run AI models directly on Qualcomm’s QCT6400 IoT chip.

○ Digital twin-based simulation: Replicates real environments in a virtual space to generate large volumes of physics-based synthetic data and then trains models in parallel on synthetic and real data to reduce error rates.

○ General-purpose foundation model + engineering: Builds on a general-purpose foundation model and tailors it through fine-tuning and engineering to on-site conditions and domain‑specific requirements to optimize performance.

○ Model compression and optimization: Adjusts performance to model size and hardware resources and flexibly replaces or combines models depending on the objective.

3. Industry Use Cases

○ Agriculture: An autonomous pesticide sprayer uses GPS and voice recognition to avoid obstacles, has secured a mass-production contract in Korea for 100 units, and has successfully entered the Indonesian market.

○ Defense: Unmanned reconnaissance robot teams perform high-risk missions, including mine detection, battlefield reconnaissance and access to hazardous areas.

○ Home Appliances: An autonomous, voice-enabled air purifier recognizes commands even in noisy environments and navigates to the requested room.

○ Public Services: An accessible kiosk automatically adjusts the screen height when it detects a wheelchair approaching and provides multilingual and voice-based interactions.

4. Global Landscape and South Korea’s Strategy

The United States prioritizes high-value, high-volume production with cloud-centric deployments for large-scale manufacturing. China focuses on low-cost production and entertainment-driven markets. South Korea is well-positioned to pursue a niche-market strategy of pairing on-device AI with regionally specialized industries—automotive, shipbuilding, agriculture and tourism—to build a differentiated competitive edge.

5. Technology and Industry Recommendations

○ Data Strategy: In a highly regulated domestic environment, leverage virtual environments and self-training to rapidly strengthen competitiveness and build on-site datasets using digital platforms tailored to regionally specialized industries.

○ Industrial Priorities: Focus on industries with long value chains (e.g., automotive, shipbuilding) and on sectors where safety and efficiency are paramount (public sector, defense, agriculture).

○ Collaboration Model: Build close partnerships between robotics and AI companies for hardware control, navigation and the development of perception and decision-making models.

○ Regional Specialization: Tailor AI applications to local industry structures and further refine custom datasets—for example, in Incheon (biotech, ports and aviation) and Jeju Island (agriculture and tourism).

6. Long-Term Outlook for Physical AI

Over the next one to two years, on-site collaboration between robots and AI is expected to see broad adoption, and within 10 years it is projected to replace physical labor across industries while creating new jobs. Physical AI is expected to establish itself as a key technology for national competitiveness.