Samsung Opens a Dedicated Physical AI Lab, Putting a KAIST-Trained Roboticist in Charge of Its Humanoid Brain

When Samsung Electronics created its RX (Robotics eXperience) Business Promotion Office on July 21, 2026 — a division that reports directly to the CEO and sent the company’s shares up roughly 6% on the day — the announcement read like classic conglomerate reorganization language: integrate strategy, accelerate commercialization, secure future growth engines. What it didn’t reveal was how quickly the org chart would fill in underneath. On August 18, Korean outlet Edaily reported that Samsung has now begun operating a dedicated Physical AI Lab under the RX umbrella, and the details of who runs it and what it builds say a lot about how seriously Samsung is taking the “brain” problem in humanoid robotics.

What the lab actually is

The Physical AI Lab succeeds the “Future Robot AI Group” that sat inside Samsung’s now-restructured Future Robot Promotion Team. Its remit is the AI and control technology that humanoids need to walk, balance, and manipulate objects in real-world environments — the layer that turns a collection of actuators and sensors into a machine that can be useful outside a lab demo.

The research agenda is specific and telling: full-body control, locomotion, and object manipulation, plus the two learning paradigms that dominate modern robotics — reinforcement learning and imitation learning, which let robots learn behaviors from human instructions or demonstrated actions rather than hand-coded rules. Edaily reports the team is already working on concrete capabilities such as making a humanoid perform full-body movements in response to human voice commands, and dynamically adjusting posture and gait to carry objects of different weights. Some of this research has reportedly been validated in real-world environments using a Chinese-made Unitree humanoid robot — a pragmatic choice of test hardware that underscores how far ahead Chinese platforms currently are as research vehicles.

A roboticist’s lab, run by a roboticist

The lab director is Koo Dong-han, a Samsung senior researcher with a Ph.D. in mechanical engineering from KAIST — the talent pipeline that has supplied much of Korea’s robotics industry. Koo spent time on Hyundai Motor’s robotics team, where he worked on wearable robots and walking robots, before joining Samsung Research in 2020. That pedigree matters: Samsung is not handing its core robotics AI program to a generalist software manager, but to someone who has spent a career on the control-theory end of the field, where the gap between simulation and reality is won or lost.

Buy versus build, resolved in favor of both

The most important context for the new lab is Samsung’s split strategy on robotics technology. On the “buy” side, Samsung acquired a 14.7% stake in Rainbow Robotics in 2023 for KRW 86.8 billion, then exercised options in late 2024 to raise its holding to roughly 35%, becoming the Korean robotics maker’s largest shareholder. Rainbow gives Samsung humanoid platforms, hardware engineering, and the RB-Y1 dual-arm robot. On the “build” side, the Physical AI Lab and a sibling “Hand Lab” — which researches dexterous robotic hands, arguably the hardest unsolved hardware problem in the field — develop core technology in-house.

The RX office that houses both labs is led from above by a Robotics Strategy Team under Executive Vice President Lee Dong-geon, another Hyundai Motor Group veteran who previously oversaw robotics strategy there. Samsung has also recruited outside academia: Professor Kim Hyun-jin of Seoul National University, an expert in autonomous robot systems, and Professor Kim Eui-gyeom of Ajou University, a specialist in robotic hands and manipulation.

Geographically, Samsung plans to use its Seoul R&D Campus in Umyeon-dong as the primary research hub, while establishing a “Data Factory” at its Gumi facility to manufacture the training data that modern robot learning consumes in enormous volumes. That Gumi investment sits inside a reported ~19 trillion won (about $13 billion) commitment to build humanoid robot manufacturing facilities and data infrastructure — numbers that position Samsung’s robotics push as a line item comparable to a major fab project, not a skunkworks.

Why now, and what it signals

Three currents are converging. First, the industry consensus has shifted: the binding constraint in humanoid robotics is no longer just hardware cost or actuator performance, but the AI stack — control policies, manipulation skills, and the data pipelines to train them. Physical AI, a term NVIDIA has popularized, is where the differentiation will happen. Second, Samsung has publicly committed to showing concrete humanoid progress by 2028, and reports suggest it has even built a confidential humanoid prototype outside the RX organization using motor technology borrowed from its home-appliance division — a classic Samsung move that exploits its unmatched vertical breadth in electromechanical components. Third, the competitive clock is running: Chinese automakers and robot makers are pivoting to humanoids at scale, Tesla continues to iterate Optimus, and Korean rival landscape leaves little room for a slow, purely outsourced approach.

CEO Noh Tae-moon, who heads the DX Division and personally championed the RX office, framed the reorganization at last month’s Galaxy Unpacked event in London: “This is an organization designed to transform the pioneering development and investment in the robotics field that we have been pursuing into full-scale commercialization.”

That sentence is the whole strategy in one line. Samsung has spent three years optioning robotics capabilities — the Rainbow stake, academic hiring, appliance-grade motors, and now a dedicated AI lab. The Physical AI Lab is the clearest signal yet that the company intends to own the software core of its humanoids outright, rather than rent it. For a firm that watched its memory and display businesses reward exactly this kind of vertical internalization, the logic is familiar: control the hardest layer, and the rest of the stack follows.

Whether a consumer or commercial Samsung humanoid materializes on the 2028 timeline will depend less on hardware — Samsung builds hardware as well as anyone alive — than on whether Koo Dong-han’s lab can close the gap between imitation-learning demos in simulation and reliable manipulation in a real kitchen, warehouse, or care facility. That is the same unsolved problem everyone from Tesla to Figure to Unitree is charging at. Samsung has now formally entered the race, with its balance sheet and its org chart aligned behind it.