Artificial intelligence (AI) is rapidly transitioning from generative capabilities to understanding and interacting with the physical world. This new era sees AI engaging with real environments across various sectors, including manufacturing, logistics, defense, healthcare, and public services. Physical AI cannot be achieved through a single technology; it requires a combination of AI that understands and judges reality, a world model that predicts the future based on real data, efficient computing, and devices that can act.
The competition is not merely about creating better AI models or superior robots. The key to success lies in how well AI models, data, semiconductors, computing, and devices are interconnected, and how quickly these connections can address industry challenges and yield market results.
The 'Physical AI Alliance' was founded on this understanding. Established last year and expanded this year, the alliance comprises three divisions: full-stack, vertical industry bridge, and foundational governance. Each division has distinct roles in technology, industry, and governance, but the ultimate goal is to connect their capabilities to create a unique physical AI ecosystem.
The full-stack division's primary focus is on expanding a universal physical AI foundation model. Creating separate AIs for each robot and task limits scalability; thus, securing a foundational AI model that can be used across various robots and environments is crucial for enhancing our physical AI competitiveness. Google DeepMind has demonstrated the potential of universal models with its 'Gemini Robotics 2,' which performs various tasks across different robots.
The second focus is on specialized infrastructure for physical AI. The learning and inference requirements for physical AI differ significantly from those of large language models. During the learning phase, which involves processing large amounts of simulation and sensor data, storage and network capabilities can become bottlenecks. In real-time inference, minimizing latency and power consumption is critical. By defining and validating a 'physical AI reference infrastructure' that combines domestic neural processing units (NPUs) and AI models, our companies can enter the global market with verified full-stack solutions rather than individual components.
The alliance's role is to identify challenges needed by the industry and connect them with national research and development efforts, ensuring that the results return to the field through collaboration between businesses and research institutions. For the 'Physical AI' initiative, one of the government's three mega-projects, to yield national results, close cooperation between the government, private sector, and research community is essential. This is why the alliance must serve as a vital link between technology and industry.
So, where do we currently stand? In individual fields such as AI models, semiconductors, manufacturing, and robotics, we possess world-class technologies and companies. Domestic physical AI company RealWorld demonstrated its potential by winning first place in the foundation model category at the 'Nebius Robotics Awards' last year. The world's leading manufacturing sites generate vast amounts of data daily, and rapidly growing robotics companies are valuable assets. However, excelling in individual technologies and integrating them into a cohesive system for market launch are entirely different challenges. While our connectivity capabilities are still insufficient, achieving proper integration is feasible.
NVIDIA is establishing its ecosystem by linking models like GR00T with computing and world model platforms like COSMOS. Following the same path as competitors will not yield success. Our competitiveness lies in consolidating our strengths. The alliance is continuing its collaboration through the full-stack division, which connects technologies from AI models to data, computing, and robotics; the vertical industry bridge division, which links industry demands with technology; and the foundational governance division, which establishes common standards and regulations. As the head of the full-stack division, I aim to gather the capabilities of academia, industry, and research to ensure that developed technologies translate into tangible industrial outcomes.
In the era of physical AI, a collaborative ecosystem is more powerful than individual advancements. To realize the goal of 'Korea as a leader in physical AI,' we must connect our respective technologies and validate them together to enhance our competitiveness in the global market. I hope the Physical AI Alliance will serve as a strong foundation to unite Korea's technology and industry.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.