[Edaily Reporter Shin Yeong-bin ] Samsung is accelerating its humanoid research in the United States. SamsungElectronics(005930)is developing robot artificial intelligence (AI)-based technologies, while Samsung SDS (SAMSUNG SDS CO., LTD.(018260)) is expanding its AI capabilities for manufacturing site operations and monitoring, broadening the company’s research front in physical AI.

According to the information technology (IT) industry on the 1st, SamsungElectronics’ U.S. research and development (R&D) subsidiary, Samsung Research America (SRA), recently established a “Data Efficiency” group within its Robotics Intelligence Lab and has begun recruiting a head to lead it.

The Data Efficiency Group researches technologies to improve learning performance even with limited robot data. Key research areas include simulation-based learning and evaluation, first-person video learning, and teleoperation data augmentation. The group focuses on reality-simulation linked learning technology, which involves transferring real-world data to a virtual environment for training and then applying the results back to actual robots.

The internal organizational structure of the SRA Robotics Intelligence Lab has also been revealed in detail. In addition to the Data Efficiency Group, the lab is building its team around four core groups: Inference, Precision Manipulation, and Robotic Systems.

The Inference Group is responsible for technologies that enable robots to understand their surroundings and plan complex tasks. The Precision Manipulation Group focuses on the ability to handle objects stably using vision and tactile feedback, while the Robotics Systems Group is tasked with integrating and validating AI technologies on actual robots.

The entire lab is led by robotics expert Chris Hauser, a professor of computer science at the University of Illinois. Hauser has consistently emphasized learning efficiency over the sheer volume of data in robot AI development. This year, he has set a goal of achieving state-of-the-art research results using less than 100 hours of real-world robot data.

Samsung SDS is also expanding its research into Physical AI. Samsung SDS America recently began recruiting new staff to conduct research on real-time video understanding and Physical AI. One of the core research projects involves AI that analyzes videos of humanoids performing tasks in real time to detect incorrect actions or abnormal situations.

The local AI Science Lab is also conducting research that involves directly operating real humanoids to collect data via teleoperation, as well as training and deploying the open-source RFM model. Previously, the company had also been recruiting personnel in robotics learning fields such as embedded AI and Vision-Language-Action (VLA) models.

Samsung SDS views the manufacturing floor as the ultimate application area. The company is reportedly set to validate the relevant research technologies within Samsung’s global manufacturing environment and demonstrate them in pilot projects on actual production lines. Going forward, the company plans to expand this into multi-robot systems where various types of robots operate together.

Last July, SamsungElectronics launched the RX Business Promotion Office, which reports directly to the CEO, and announced plans to expand its global robotics research hubs in the U.S., China, and Japan. As this initiative aims to consolidate scattered capabilities—ranging from mid- to long-term robotics strategies to the development and commercialization of core technologies—interest in its U.S. research strategy is also growing.

Hauser believes that the commercialization of general-purpose robots is not a race that will be won in the short term. He recently expressed caution regarding the competition over the sheer volume of data on social media, stating, “Existing robot foundation models (RFMs) alone are not sufficient.”