Samsung Electronics (005930.KS) has preemptively secured artificial intelligence technology for controlling humanoid robots—essentially the robot's brain. While the company has not yet officially unveiled a completed humanoid prototype to the public, analysts say Samsung is laying the groundwork for building its own robotics ecosystem by first solidifying its software capabilities, the key to commercialization.
According to industry sources on the 12th, Samsung Research (SR) recently announced "Shallow-π," a technology that reduces the computational steps of AI models for robot control to one-third of previous levels. The technology features "knowledge distillation," a technique that compresses the core intelligence of large-scale AI models into smaller models.
This advancement more than doubled the robot's situational judgment speed from 8Hz to 17.2Hz. With over 17 consecutive decisions possible per second, the system now has the responsiveness to instantly react to unexpected obstacles and stop without accidents. Notably, by overcoming the structural limitations of conventional robot AI that required cloud servers for massive data processing through its proprietary on-device AI technology, Samsung is considered to have cleared a major hurdle toward commercialization.
Field verification was also thoroughly conducted. The system demonstrated normal operation on Nvidia's latest robot platform "Jetson Orin" and humanoid-dedicated chip "Jetson Thor" environments. Samsung adopted a method that processes everything from sensor input to motion output in a single flow using a unified Vision-Language-Action (VLA) model. As a result, the system achieved a 95% success rate in ultra-precision water hose insertion tasks requiring less than 1mm error tolerance, and also accomplished manipulating 46 movements of dual arms and robot hands with 22 degrees of freedom in just 40 milliseconds.
Behind Samsung Electronics' focus on securing software capabilities before revealing the physical form lies its mid-to-long-term roadmap of "complete transition to AI autonomous factories by 2030." Rather than attracting attention with flashy service robots, the strategy is to deploy robots preemptively in harsh manufacturing environments with high temperatures and noise to substantially improve production efficiency. The company appears set to introduce digital twin simulations for core processes from material receiving to production and shipping, and systematically deploy operating bots and assembly bots.
This vision was partially revealed through DX Division President Roh Tae-moon's remarks at CES 2025 in January and earnings conference calls. The scenario involves accumulating physical AI data through challenging tasks in industrial settings, securing mechanical safety, and then gradually expanding into the business-to-consumer (B2C) market. If this lightweight on-device AI is integrated into the control technology of partners such as Rainbow Robotics, in which Samsung has invested, commercialization timing is expected to be accelerated further.
In the global humanoid robot market, projected to grow to $4.4 billion (approximately 6.34 trillion won) next year, Chinese companies are currently leading in physical unveilings and external design competition.
Samsung is countering this by recruiting top global talent with exceptional compensation packages to secure technological advantages. Samsung Research America (SRA) recently entered the robotics talent race directly, offering annual salaries of up to $365,650 (approximately 530 million won)—exceeding senior researcher levels in the memory sector. The company is strengthening its R&D lineup by successively recruiting top-tier talent including Professor Chris Hauser from the University of Illinois and experts from NASA and Honda.
Large-scale investment has also been announced. Samsung Electronics is expected to accelerate humanoid development this year based on record R&D investment exceeding 100 trillion won.
