“Countries around the world are investing massive amounts of capital in manufacturing and other fields, and are achieving results in physical AI through AI applications in humanoids, rockets, high-speed rail, military robots, and drones. “Our country also needs to achieve independence in core physical AI technologies through strategic expansion centered on various industrial bases such as smart cities, healthcare, and national security in addition to manufacturing and logistics through continuous large-scale investment.”

Lee Hae-soo, a senior researcher at the AI Policy Research Lab of the Software Policy Research Institute, gave a presentation on the topic of “Physical AI Development Trends and Related Policy Implications” at the “2025 Embedded AI Trend Forum” hosted by the Embedded Software and Systems Industry Association (Chairman Lee Chang-yeol, hereinafter referred to as KESSIA) in the main auditorium of the Korea Conference Center on the 16th. He stated that physical AI is a next-generation technology that determines industrial competitiveness and national security, and that preemptive response is urgent.

Senior Researcher Lee Hae-soo said, “Physical AI, which is receiving attention from domestic and international industries and academia, refers to an intelligent system that continuously performs recognition, inference, planning, and action and operates autonomously in the physical world.” He added, “It is moving beyond generative AI that handles text and images to the stage of ‘physical AI’ that permeates hardware such as actual robot arms, self-driving cars, and drones, and its areas of application are rapidly expanding to manufacturing, logistics, healthcare, and smart cities.”

He continued, “At a time when AI agents were autonomously operating software environments, NVIDIA CEO Jensen Huang declared that ‘the next frontier is physical AI,’ and continuously disclosed the GPU chip roadmap.” He added, “According to global market research firm Grand View Research, the physical AI market is expected to grow from $12.7 billion in 2023 to approximately KRW 168 trillion by 2030, at a CAGR of 38.5%.”

The core technology of physical AI is multimodal learning (visual and audio)The examples include sensors and computer vision that enhance the accuracy of physical environment recognition with AI algorithms for autonomous decision-making in complex environments (3D spatial data) and robot-specific models (VLA), and sensor fusion based on lidar, radar, and events, as well as network and edge computing with edge servers and MPUs (mobile processing units) that reduce cloud dependency and enable real-time on-site computation.

In particular, actuators, which are called the 'muscles' of physical AI, generate movements using motors, reducers, and encoders. Recently, smart actuators with integrated AI control have emerged, enabling real-time environmental adaptability and reduced design complexity, while enabling advanced gripping such as the Allegro hand from MetaWork Robotics.

Looking at major application cases, in humanoids, Tesla's 'Optimus' and Figure AI robots are in the manufacturing and service field demonstration stage, and in autonomous driving, Tesla, Google Waymo, BYD, etc. are accelerating the development of urban robot taxis by integrating FSD, lidar, and AI models.

In the case of drones (UAVs), they are used for autonomous flight, night photography, and precision pest control in the military, agriculture, and public safety, and in the case of AMRs and AGVs, SLAM-based autonomous movement logistics solutions are attracting attention in irregular warehouses, hospitals, and hotels.

Meanwhile, there are technological and social limitations and challenges, such as the need to strengthen adaptability to the highly uncertain real environment due to the gap between simulation and reality, and the need to develop high-density lithium cells and low-power semiconductors in relation to energy efficiency and batteries.

There are also ethical and legal issues to be addressed, including excessive costs, unclear responsibilities, and privacy, human dignity, and safety regulations.

Meanwhile, countries around the world are spurring the development of physical AI.

The United States announced the revision of the 'Robotics Roadmap' and R&D projects that began in 2009, and is supporting approximately 300 projects.

China has officially stated ‘Made in China 2025’ and ‘physical AI’ at the National People’s Congress, and is controlling parts and component networks.
<br /> The EU and Japan are pursuing AI regulations and ethical principles in parallel, and Japan is pursuing a recovery strategy with its strengths in industrial robots.

Korea is also pursuing three strategies to increase the size of the K-robot industry by more than four times from the current 5.6 trillion won to 20 trillion won by 2030: △ Strengthening the three core competitiveness of robots: technology, human resources, and companies; △ Full-scale expansion of K-robots; and △ Creation of a robot-friendly environment. The K-Humanoid Alliance is also being promoted.

However, in the case of our country, various key components that can be included in physical AI have not been self-sufficient, so it was determined that a lot of effort is needed to localize these parts.

Senior Researcher Lee Hae-soo advised, “In terms of policy implications for physical AI, we should actively pursue the following: continuing the national physical AI strategy, strategic expansion centered on core industrial bases, establishing a ‘Phisical AI Strategy Committee,’ investing in domestic production of core components and software, supporting the establishment of test beds and alliances, strengthening human resource development, and expanding international cooperation.”