[No Filter Robot] Ryu Seokhyun's Winning Move... The Path Forward for Korean Humanoids Is 'Data'

Korea Institute of Machinery and Materials Unveils Robot Task AI
Securing Data to Enhance Task Performance with AI Is More Important Than Hardware
President Ryu Seokhyun: "We Should Compete in Task Performance, Not Just Mobility"

"Bring me the banana on top of the washing machine"

When Ko Dooyeol, Senior Researcher at the Artificial Machinery Research Lab of the Korea Institute of Machinery and Materials, gave the command, the robot recognized its surroundings and began moving toward the washing machine. Standing in front of the place where the banana was located, the robot extended its arm and picked up the banana. It then returned to its original spot and placed the banana on the dining table.

A robot equipped with robotic work AI developed by the Korea Institute of Machinery and Materials is moving to pick up a banana after receiving the command "Bring the banana." Photo by Paek Jongmin, Tech Specialist

In the subsequent demonstration, the command "Sort the trash for recycling" was issued. The robot picked up the trash on the table, moved to the recycling bin, and sorted the waste by material—placing plastics and cans into their respective bins.

On March 12, the Korea Institute of Machinery and Materials unveiled its robot task AI to the press at its Daejeon headquarters, showcasing how robots can learn human tasks to perform a wide range of functions.

This research is being carried out as part of the project "Development of Core Technologies for the RoGeTA Framework for Implementing Diverse Everyday Services with Robotic General Task AI." The goal is to develop robotic intelligence technologies to support daily tasks—an essential capability for humanoid robots designed to assist with household chores.

The objective is for robots to learn from human demonstrations, using this data to perform daily tasks such as tidying up, moving objects, and sorting waste for recycling.

Sukhyun Ryu, President of the Korea Institute of Machinery and Materials (right), is explaining data collection for robot operation AI and robots. Photo by Paek Jongmin, Tech Specialist

According to the Korea Institute of Machinery and Materials, teaching "work" to robots involves three core technologies.

First, when a person demonstrates a task, Task Extraction AI converts the demonstration into data. Virtualization AI transfers actual spaces into virtual environments to test various scenarios. Task Execution AI then understands commands, calculates the task sequence, and carries out the execution.

The research team has built spaces within the institute that simulate real home environments to collect data. In these recreated bedrooms, kitchens, and living rooms, robots are trained to perform tasks such as organizing items and cleaning.

A representative from the institute explained, "We constructed experimental environments centered on the most common daily tasks—organizing, cleaning, and kitchen work. We collect data in these spaces and conduct research so that robots can perform tasks in real-world environments as well."

The developed robot task AI can be widely applied to various operations, including service work in homes and offices, shelf organization in retail stores, and picking and sorting tasks in logistics sites. Going forward, the plan is to further expand the range of tasks robots can perform and strengthen their adaptability to changes in space and objects, thereby enhancing their usefulness in real service environments.

The data collection system for robot learning was also presented to the press. In this system, a researcher directly operates the robot to perform tasks, and the humanoid-type robot learns from this data.

Although still in its early stages, the data collected in this way is being used to teach robots how to perform tasks.

To this end, the team is also working on establishing an "Open Data Factory." In this project, several robots are deployed in a space of about 1,000 square meters to collect data in various environments, which is then analyzed and processed for use by researchers and companies. This project is part of the AI Humanoid Global Top Strategy Research Group led by the Ministry of Science and ICT.

Park Chanhoon, head of the research group, explained, "We are securing a variety of robots and collecting data."

He further stated, "Securing data generated in diverse situations is key for robots to function properly. The collected data will also be made available to domestic researchers and companies to help advance the robot industry ecosystem."

Recently, competition in the humanoid robot industry has been intensifying. Global robotics companies are unveiling demonstrations of humanoid robots performing tasks in home environments, vying for leadership in the field.

Just before the demonstration by the Korea Institute of Machinery and Materials, U.S. startup Figure AI released a video of its humanoid robot performing household organizing and object-moving tasks. Its appearance—walking on two legs and organizing objects with two arms—resembled that of a human.

Electric vehicle manufacturer Tesla is developing the humanoid robot "Optimus" with the goal of automating its factories. In China, companies such as UBTECH and Unitree are also entering the humanoid robot development race.

Boston Dynamics, a subsidiary of Hyundai Motor Company, drew significant attention by showcasing the potential of physical AI with its "Atlas" humanoid. However, Ryu Seokhyun, President of the Korea Institute of Machinery and Materials, assessed that Korea has had a somewhat late start in the humanoid race.

He said, "For a while, there was a perception in Korea that the business model for humanoids was uncertain, and there was insufficient national support. Humanoids are among the most complex mechanical systems on Earth, making it harder to realize them quickly than many expect."

He continued, "Mobility and dexterity are separate challenges," emphasizing that the opportunity to catch up has not completely disappeared.

President Ryu explained robot technology by dividing it into two domains: mobility and dexterity.

"Excelling at kung fu is a matter of mobility, while picking up a banana or sorting waste is about dexterity and task performance. These are completely different domains."

He added, "There is currently no robot that possesses both of these capabilities; in the areas of task performance and dexterity, there are still opportunities for us."

President Ryu especially highlighted the importance of data-driven learning infrastructure in the robotics competition.

He stated, "To develop robots that function properly, it's ultimately critical to obtain data from diverse environments and use it for training." This is the reason the institute is prioritizing the acquisition of robot data.

This content was produced with the assistance of AI translation services.