The competitive landscape of artificial intelligence is entering a new phase. While generative AI and large language models (LLMs) have dominated the industry over the past several years, the focus is increasingly moving beyond “talking AI” toward “Physical AI” — systems capable of perceiving, reasoning and acting within real-world environments.
This transition formed the backdrop to the 50th anniversary forum hosted by the Electronics and Telecommunications Research Institute (ETRI) on May 28. Held under the theme, “The Ultimate Weapon of AI: The Road to Physical AI,” the forum carried significance beyond a conventional research showcase, serving instead as a platform to discuss the future direction of AI, robotics and computing convergence.
Physical AI differs fundamentally from conventional chatbots and generative systems. Rather than remaining confined to text or image generation, it integrates with robotics, autonomous driving, smart factories and digital manufacturing to make decisions and perform actions in physical environments. In effect, AI is evolving from generating responses on screens to understanding and interacting with the real world.
After Generative AI, the Competition Turns to ‘Acting AI’
Across the global AI industry, competition surrounding humanoid robots, embodied AI, world models and Robot Foundation Models (RFMs) is accelerating rapidly. This is no longer merely a race to build better robots; increasingly, it is a contest over how effectively machines can understand, interpret and learn from physical environments.
ETRI reflected this trend by identifying “robot intelligence” as a strategic priority within its AI robotics agenda. The institute said future value creation in robotics will depend less on hardware itself and more on intelligent systems, signaling plans to strengthen related research efforts.
The implications extend beyond technology into industrial structure. Earlier generations of robotics competition centered on precision motors, sensors and mechanical performance. Today, competitive advantage is increasingly shaped by who possesses larger behavioral datasets and who can train more capable general-purpose robot intelligence systems.
Presentations delivered during the forum mirrored this structural transition. Kim Seung-hwan, head of the Expert AI Group at LG AI Research, discussed industrial transformation and corporate strategy under the theme “The Era of Physical AI and Our Industrial Direction.” Professor Park Jong-woo of Seoul National University examined the prospect of robot-driven industrial transformation in his presentation, “Industrial Transformation Led by Physical AI.”
RFM and Robot Intelligence Become New Competitive Standards
Among the forum’s discussions, ETRI’s Meta RFM-based strategy drew particular attention.
If LLMs functioned as foundational models for language-based AI, RFMs are increasingly viewed as foundational architectures for Physical AI, capable of learning spatial awareness, behavioral planning, human-robot interaction and task execution. The implication is significant: the future of robot intelligence lies not in simple machine control but in building general-purpose AI systems capable of understanding the physical world.
Yu Won-pil, president of ETRI’s AI Creative Research Laboratory, presented several long-term strategic priorities, including Meta RFM-based adaptive robot intelligence, self-growing AI robot ecosystems and the development of sovereign robot data.
Among these, the concept of “sovereign robot data” carries broader national implications.
If the LLM era revolved around text and language datasets, the Physical AI era may increasingly depend on manufacturing-site information, human motion data, robot behavioral data and industrial operational datasets. In other words, AI competition is expanding beyond model performance toward a more fundamental question: who owns and trains on real-world data.
This transition also reshapes the meaning of data sovereignty itself. The debate is moving beyond language and search-based digital sovereignty toward sovereignty over industrial and physical-world data.
The Age When AI Safety and Standards Become Infrastructure
Physical AI brings not only opportunity but also new responsibilities.
This explains why Kim Myung-joo, director of ETRI’s AI Safety Research Center, addressed “AI Robot Ethics” during the forum. Errors produced by chatbots may result primarily in misinformation, but failures involving robots and autonomous systems can lead directly to industrial accidents, safety risks and legal liability.
Yu stressed that ethical and governance principles should apply throughout the full AI robot lifecycle—from design and development to testing, deployment and operation. He also introduced a proposed standardization framework categorizing robot intelligence from Level 1 through Level 5.
Such efforts represent more than academic classification. They can be interpreted as attempts to build the “trust infrastructure” necessary for Physical AI commercialization. Industrial deployment of AI robotics will ultimately require not only performance verification but also accountability and safety validation.
The panel discussion that followed, focused on “Global Top Strategies for Physical AI and ETRI’s Role,” further explored pathways for strengthening Korea’s AI robotics competitiveness through cooperation among industry, academia and research institutions.
The significance of ETRI’s forum extends beyond the event itself. If generative AI marked an era in which machines learned digital language, Physical AI signals a transition toward systems capable of learning the rules and behaviors of the physical world.
Ultimately, the center of AI competition is expanding from screens into real space. Physical AI is emerging not merely as a new robotics technology, but as a strategic infrastructure shaping the next stage of the AI industry.
