Startup in South Korea teaches humanoid robots human skills
David Park, with his head, chest, and hands equipped with body cameras, skillfully folded a banquet napkin, a task he has performed thousands of times over his nine-year tenure at the prestigious Lotte Hotel Seoul. Every action he takes is recorded in a database that aims to eventually instruct a robot to replicate those actions. The hotel chain represents one of numerous partnerships that South Korean artificial-intelligence startup RLWRLD (pronounced “real world”) is pursuing to establish a comprehensive repository of human expertise. This initiative involves gathering insights from skilled workers across various industries to construct AI systems for robots that may soon be deployed in industrial settings and residential environments. The study gathers comparable data from logistics personnel at CJ, focusing on their methods of gripping, lifting, and managing goods within warehouses, as well as from employees at the Japanese convenience store chain Lawson, monitoring their strategies for organizing food displays. The objective is to develop an AI software layer capable of operating in robots across various factories and work sites in the forthcoming years, with the possibility of later extending its application to residential environments. Engineers at RLWRLD emphasize that achieving the dexterity of human hands is a crucial objective, underscoring their belief that humanoid machines will be pivotal in advancing the sector. “I’ve been doing this about once a month,” said Park, one of approximately 10 members of Lotte Hotel’s food and beverages team being equipped to document their techniques.
After meticulously folding the napkin into a compact, layered square, Park attended to the wine glasses, knives, and forks positioned in a corner of the banquet hall, while colleagues organized for the actual service in close proximity. He expressed a mild grievance to an engineer regarding the excessive tightness of the cameras on his hands. RLWRLD stands out in a burgeoning cohort of South Korean high-tech enterprises and manufacturers vying for dominance in the nascent yet intensely competitive global arena of “physical AI.” The term denotes machines that are outfitted with artificial intelligence and sensors, enabling them to perceive, make decisions, and operate within real-world environments with a certain level of autonomy, thus advancing beyond traditional factory robots that are tailored for repetitive functions. While it remains uncertain if these machines will completely fulfill the expectations of revolutionizing industries, they are pivotal to South Korea’s aspirations to capitalize on its semiconductor and manufacturing capabilities to establish itself as an AI leader. The competitive landscape is formidable, characterized by US technology leaders such as Tesla and a surge of Chinese enterprises investing heavily in humanoid and various AI robotics. Similar to how chatbots like ChatGPT and Gemini are trained on extensive collections of internet text, AI robots also necessitate substantial data regarding human behavior to effectively perform complex physical tasks. South Koreans may face challenges in the chatbot arena, where proficiency in English provides significant advantages to US companies. However, they perceive greater opportunities in physical AI, leveraging their substantial pool of skilled labor in manufacturing and related fields that can contribute to the training of robotic systems.
Last month, the government unveiled a project valued at USD 33 million aimed at capturing the “instinctive know-how and skills” of “master technicians” in a database designed for AI-driven manufacturing. The initiative seeks to enhance productivity through robotics while addressing the challenges posed by an aging and diminishing workforce. RLWRLD, which last week unveiled its robotics foundation model, an AI system for robots, anticipates that industrial AI robots will be deployed at scale around 2028, a timeline consistent with projections from major businesses. Hyundai Motor intends to deploy humanoids developed by its robotics division, Boston Dynamics, across its global manufacturing facilities in the forthcoming years, commencing with its Georgia plant in 2028. Samsung Electronics, a leading player in the semiconductor industry, has announced its intention to transform all manufacturing facilities into “AI-driven factories” by the year 2030, incorporating humanoids and task-specific robots throughout its production lines. According to Billy Choi, South Korea boasts a highly developed manufacturing sector with a concentrated emphasis on humanoids designed specifically for these industries. South Korea’s initiative in artificial intelligence has raised concerns among labor organizations, who worry that automation may lead to job displacement and undermine the skilled workforce that has historically provided the nation with a competitive advantage—an asset that is now deemed crucial for its transition to AI.
In January, Hyundai’s union cautioned that the rise of robots might lead to a “employment shock.” In response, President Lee Jae Myung delivered an unusual critique, characterizing AI as an unstoppable “massive cart” and urging union members to adjust to the rapid changes that are “coming faster than expected.” Kim Seok stated “Mastery of skills is ultimately a human achievement – even if AI can replicate existing abilities, the continuous development of craft will remain fundamentally human.” He indicated that extensive robot deployments could jeopardize the “pipeline” for skilled labor and called on the government and employers to collaborate with workers regarding AI to secure their support and alleviate job-related anxieties. Humanoids created by companies in the United States and China have demonstrated remarkable physical capabilities, including long-distance running. Hyemin Cho, responsible for business strategies at RLWRLD, stated that the capacity to execute intricate tasks manually will be pivotal in assessing the applicability of humanoids across various industrial environments and domestic spaces. “Capturing motion data in real-world settings is extremely important, and the quality of that data matters greatly,” she stated.
Following the transformation of worker footage into machine-readable data, RLWRLD’s engineers enhance the process by re-executing those tasks equipped with cameras, VR headsets, and motion-tracking gloves. The data serves to train test robots, frequently directed by RLWRLD “pilots” employing wearable devices. According to Song Hyun-ji from the robotics team, the process meticulously captures intricate details, including joint angles and the magnitude of force exerted. One of RLWRLD’s laboratories is situated in a disorganized suite on the 34th floor of the Lotte Hotel. Worn carpets lie obscured beneath a chaotic array of wires and technological equipment. Poles equipped with infrared laser readers are positioned in the corners. Underneath a chandelier, a rare remnant of the room’s previous opulence, a wheeled robot equipped with black, humanlike metal hands traverses back and forth, emitting a low mechanical whir. In a recent demonstration, the robot, under the guidance of engineers, carefully lifted and positioned cups at a minibar, inadvertently toppling a dish at one moment. The recent test footage from the company reveals a more sophisticated system: a humanoid adeptly opening a box, placing a computer mouse within, closing it, and subsequently positioning it on a conveyor belt. Most robots, including Boston Dynamics’ Atlas, utilize task-specific hands, such as two or three-fingered “grippers.” RLWRLD is part of a select cohort of firms engaged in the development of AI technologies aimed at creating five-fingered hands that replicate the nuances of human touch.
Choi noted that although five-fingered designs may not consistently align with factory requirements, they could become essential as robots transition into domestic environments, necessitating more intimate interactions with humans. According to Cho, hospitality workers contribute essential training data for machines that are learning specific or intricate tasks—abilities that could further enhance their application in industrial environments. While present-day humanoids require several hours to complete the cleaning of a guest room, a task that human workers accomplish in approximately 40 minutes, Lotte Hotel anticipates that robots will be prepared for cleaning and other behind-the-scenes functions by 2029. It also intends to offer robot rental services for the hospitality and other service sectors, with a possible extension to residential applications. “If you look at the entire process of preparing for an event in back-of-house areas, we think humanoids might be able to take over about 30 per cent to 40 per cent of that workload,” Park said. “Replacing the remaining 50 percent, 60 percent, and 70 percent will pose significant challenges, particularly as it necessitates genuine human-to-human interaction.”
