On Tuesday 25 August 2026, South Korea’s Ministry of Science and ICT sat down at the Korea Institute of Science and Technology in Seongbuk-gu, Seoul, to take comments on a draft “K-Moonshot AI Humanoid Mission Roadmap.” No new robot walked out of that room. The useful output is a plan: pick a national K-humanoid platform inside three years, and start a public contest in 2027 to decide who gets to build it.
That is industrial policy with an audition attached. It is not a product launch, and it is not a spec sheet.
What Tuesday actually decided
Yeo Jun-gu (여준구), the mission’s program director, presented the draft. The ministry also walked through a new support program, the Next-X Humanoid Challenge, and appointed a Humanoid Mission Steering Committee. DigitalToday, Seoul Economic Daily, Aju Press, Maeil Business, Yonhap, MoneyToday, and Newsis all filed on the same meeting. The goal, as they quoted the ministry, is to take core-technology work — AI models and key parts — through demonstration and into mass production, using Korea’s manufacturing base, and to localise those models and parts. The prize sentence is the same in the English and Korean copy: a world-class national-team K-humanoid platform inside three years.
The Next-X Humanoid Challenge is how they intend to pick it. Starting in 2027, domestic humanoids go into an open call and evaluation. Candidates get development support. A national-level platform is chosen at the end. Maeil Business, writing the same day, framed it as a humanoid version of the Independent AI Foundation Model (Dokpamo) contest that is already picking a national AI model. The more rounds a team survives, the more support it is supposed to receive. Extra GPUs and money for data purchase are “being considered.” Those are not signed program rules. They are what the ministry is talking about while the roadmap is still a draft.
Maeil Business also said the government is leaving room for more than one winner. Humanoids come as bipeds and as wheeled machines, and the jobs are different. That is the one honest engineering sentence in the policy copy. A wheeled torso that sorts totes and a biped that is supposed to climb a stair are not the same procurement. If the contest pretends they are, the score sheet will be fiction.
The roadmap is not law yet. The ministry said it will fold Tuesday’s comments in, consult other ministries, and table the revised draft at the science and technology ministers’ meeting next month — September 2026. Until that meeting, treat “national platform” as an intention.
Yeo’s line was the same in Aju Press, Seoul Economic Daily, and Yonhap. The next wave after AI, he said, is physical intelligence that judges and moves on its own in the physical world, and the humanoid is the core platform. He added that Korea should raise a technologically sovereign K-humanoid as a national strategic asset, including to answer a shrinking population. Kim Sung-soo, who runs the ministry’s R&D Policy Office, told the room — in the wording carried by Etoday and Newsis — that a humanoid is not just a gadget. It is where AI, semiconductors, and batteries stack. He said the ministry will keep talking to people who actually build them until the roadmap can be executed.
Yeo was already the humanoid mission director. Chosun’s English report on the 27 May 2026 K-Moonshot launch, at Seoul Dragon City in Yongsan, listed him among the twelve program directors and identified him then as CEO of Daedong Robotics. K-Moonshot itself is the larger AI-for-science program: double research productivity by 2030, twelve national missions by 2035. Tuesday was one mission, not the whole tent.
Who is already on the audition list
Maeil Business named the obvious government-lab machines: KAPEX at KIST, developed with LG Electronics and LG AI Research, and Kairos at the Korea Institute of Machinery and Materials. On the company side it named LG Electronics, Doosan Robotics, and Rainbow Robotics. Those are names a reporter can see on a candidate list. They are not Tuesday’s winners.
KAPEX is the one Korean full-size humanoid with any published size I can attach to a named engineer. When KIST first showed the body, Lee Jong-won, who heads the KIST Humanoid Research Group, told Seoul Economic Daily it stands 170 centimetres and weighs in the 60-kilogram range. The same report said it can lift as much as 20 kilograms. Robot Newspaper, writing when KIST and LG announced the collaboration on 30 September 2025, said the team developed its own high-output full-body actuators and plans to put LG AI Research’s EXAONE vision-language model in as the brain. Lee later told Seoul Economic Daily the EXAONE VL mount was still ahead of them. I have not seen a KIST page that publishes whole-body degrees of freedom, joint torque, a battery capacity, a runtime, or a thermal map. Those fields are unpublished. I will not fill them in.
A separate, earlier track already sits under the same mission flag. On 18 May 2026, also at KIST, the ministry kicked off a public-private project Dong-A Science reported at 50.4 billion won through 2030, with KAPEX as the underlying machine. LG Electronics, on that account, is supposed to build 15 full-body units and five upper-body units over five years. Field trials are aimed at 2029, centered on Hallym University Sacred Heart Hospital — cleaning, recycling, ward delivery. Kim Sung-soo called that May meeting a starting point, not a finished platform. Tuesday did not replace it. Tuesday added a contest that could still pick someone else.
Kairos is thinner on paper. KIMM’s own English press release, dated 14 April 2026, says the machine made its first official appearance at the institute’s 50th anniversary in Daejeon, under the AI Humanoid Global Top Research Lab that the science ministry supports, and that it is still in an early development stage. The release does not give height, mass, degrees of freedom, payload, or battery. I am not going to invent them.
What a public contest would have to force into the open
A public contest only means something if the score sheet is public. If Next-X is going to pick a national platform, someone will have to publish the boring page: continuous joint torque, not peak; payload at the wrist with the arm outstretched, not a lab lift with the elbow tucked; winding temperature after twenty minutes of stair work; how the pack is cooled when the computer and the hip motors want the same air. A 170-centimetre, 60-kilogram claim is a size class. It is not a duty cycle. Twenty kilograms of lift, if it holds up, still does not tell you what the knee does after a shift of stair-and-carry. Those are the numbers a flagship contest would have to make someone write down.
The contest, as described, is supposed to run about three years from 2027. Winners, Maeil Business wrote, would get help commercializing and going overseas, including domestic purchase support. That is the industrial-policy half. The hardware half is still a blank form. Tuesday did not fill it in. September’s ministers’ meeting will not fill it in either, unless someone tapes a torque table to the draft.
FIRGELLI builds the linear actuators you put under a lift, a fixture, a hatch — the boring layer around a cell. Nobody in Seongbuk-gu announced one of those joints, and I will not pretend they did. What Tuesday announced is who gets to try out for the national body, and how little of that body’s engineering is public while they do it.