[Paek Jongmin's Deep Tech] Between Megaprojects and 3 Billion Won

Plugging in a Power Cord: The Greater Challenge for Humanoid Robots than Dancing
Physical AI: A Fierce 'Money War' and Race for Data
Bold Investment Needed in Top-Performing Research for Korea to Lead the World

When I visited Aegibot, a robotics company in Shanghai, China last month, what caught my eye first was the “group choreography.” Humanoid robots were dancing in line, just like K-pop idols. In a video released from the Chinese Robot Sports Competition, there were even robots that could jump higher and run faster than humans. I could truly sense the rapid evolution of humanoids.

A robot developed by the Korea Institute of Machinery and Materials is performing national calisthenics. Just a few months ago, this robot was only able to shake hands, but thanks to the continuous efforts of the research team, it is rapidly improving its capabilities. Photo by KIMM

Yet, what surprised me even more was something else: the robot data training site. Robots were imitating human movements, failing, and then trying again. There were at least thousands of robots and people at the site. I had seen robots accumulating data through repeated failures before, but this time, the scale was on another level. The CEOs who participated in the Seoul National University Artificial Intelligence (AI) CEO program were also wide-eyed with amazement. The group choreography was just a show—the real competitive edge was the data being accumulated in the background. It felt as though the Chinese company permitted the tour out of confidence, as if to ask, “Do you really think you can keep up with this?”

Dancing or running is no longer as difficult as it used to be for robots. In fact, plugging a plug into a power outlet is harder. For a human, this is a very simple action, but for a robot, it requires finding the location of the hole based on camera footage and finely adjusting the angle and force of its wrist. Even a slight misalignment leads to failure. Researchers on site say that, in the world of physical AI, plugging in a single plug is more difficult than flashy group choreography.

This paradox illustrates the essence of the physical AI race. While generative AI has grown by learning from text and images on the internet, physical AI learns by falling and failing. Every repetition—grasping, dropping, and regrasping an object—becomes a learning asset. Data does not arise on its own. Robots, components, physical space, and computing resources must come together.

This issue became even clearer as I watched the Korea Institute of Machinery and Materials’ humanoid “KAIROS” recently. In April, the robot uncertainly walked out and shook my hand. Just a few months later, it was following along with national calisthenics and dance moves. This is the result of the research team’s relentless effort. It is also understandable why the robotics community is asking for an exemption from Korea’s 52-hour workweek restriction.

However, an even bigger concern I heard on site was about funding—specifically, cash. Chanhun Park, head of the Global Top Strategy Research Division at KIMM, said they would need an additional 3 billion won (KRW 3 billion) to move to the next level. Professor Byung-Tak Zhang of Seoul National University remarked that such an amount would not be sufficient if the goal is for Korea to become the world’s leading force in physical AI.

It is not a matter of whether 3 billion won is a lot or a little. The government has designated physical AI as one of Korea’s three mega-projects for a national leap forward. Yet those on the cutting edge are worried about securing tens of billions of won to reach the next stage. If it is hard to work overnight and there is no money to run robots and generate data around the clock, nothing else matters.

China is already taking a different approach. Aegibot is creating large-scale data training sites, and Unitree has attracted massive capital through its listing. National and local governments, companies, and even the capital markets have entered the race for physical AI. Physical AI is not only a technological competition but also a “war of capital.”

Korea still has a chance. The country boasts world-leading manufacturing sites in semiconductors, automobiles, shipbuilding, and batteries. These factories can be turned into schools for robots. By gathering data on the production line and creating virtuous cycles where failed robots are retrained using domestic actuators, robot hands, and tactile sensors, Korea can foster sustained growth.

The government’s plan includes data, key components, and demonstration projects. The challenge is in execution. If these are divided among ministries and demonstrations are carried out separately, progress will be slow. Combining several projects worth billions of won will not automatically make them a mega-project.

Data, key components, and demonstration projects must be combined into a single long-term program, and bold investment should be channeled into research that yields results. Not only government funding, but also capital from private companies and the financial markets should flow into these efforts. If this is truly a mega-project, then the approach to investment must rise to the level of a mega-project as well.

This content was produced with the assistance of AI translation services.