Domestic game companies that made "Lineage" and "Battlegrounds" are joining hands with manufacturing conglomerates such as Hanwha, POSCO, and Hyundai Rotem to develop a "robot brain." The plan is to preempt the "physical AI" market by combining domestic conglomerates' manufacturing capabilities with game companies' technology for building and operating virtual battlefields, so systems can judge and move on their own in the real world. They also plan to expand into the defense AI sector for use on the battlefield.
According to the game industry on the 1st, NCSOFT's artificial intelligence (AI) subsidiary NC AI recently decided to develop a "robot foundation model (RFM)" that serves as a robot's brain with POSCO DX. NC AI will focus on advancing a VLA model that integrates vision, language, and action processing, while POSCO DX will build a Digital Twin (replicating the real space identically in virtual space) test environment based on its long-accumulated automation and operations technology. The goal is to develop a "general-purpose robot AI model" that will serve as the brain of next-generation industrial robots that understand and respond to situations on their own even in unfamiliar environments.
Physical AI is AI that operates in the real world, such as in robots, self-driving cars, factories, and logistics. Game companies are drawing attention in physical AI because of their capabilities and data for creating 3D virtual environments accumulated over years of making and running hit games.
Until now, corporations and experts have trained robot brains by having them repeatedly perform preset tasks in fixed environments to build robots that judge situations and move autonomously. But relying on this method alone creates problems, as robots cannot respond flexibly to complex, fast-changing situations. Training robots on site to cope with uncertainty through trial and error takes significant time and expense. That is why an approach in which thousands of robots simulate diverse situations in virtual environments similar to reality and turn those into training data is gaining traction.
Game companies' strength lies in their ability to build precise simulation environments needed for robot training, based on the technology accumulated while creating virtual battlefields in games, and to provide high-quality synthetic data. Massively multiplayer online role-playing games (MMORPGs), in which domestic game companies excel, unfold in virtual worlds where many users connect and act simultaneously, and in these physics-governed environments, the movements of game characters and weapons are implemented as precisely as real objects.
Game companies' capabilities in building Virtual Reality (VR) are also expected to be used to develop an "AI staff officer" that integrates and analyzes information such as battlefield maps, reconnaissance footage, and military communications to support commanders' decision-making, as well as robots to be deployed on the battlefield.
Krafton in Mar. agreed to work with Hanwha Aerospace on joint research and development of physical AI technology and to pursue the establishment of a joint venture (JV). In addition, Krafton established the Robotics research subsidiary Ludo Robotics and is accelerating the expansion of its physical AI business, cited as a next-generation growth engine, by setting up an Autonomous Driving entity with SOCAR worth 150 billion won.
Earlier, NC AI formed a consortium with Hyundai Rotem and was also selected as the final contractor for the national research and development project "physical AI-based integrated simulator and modular robot system" commissioned by the Agency for Defense Development (ADD). The purpose is to enhance the completeness of manned-unmanned combined weapons systems that organically control diverse, multiple unmanned robots on future battlefields. In this project, NC AI will lead the development of the "world model," a core technology needed to implement the robot foundation model.
Even if robots have learned through countless trials and errors in virtual environments, the simulation-to-reality gap—malfunctions in the face of subtle real-world variables—is the biggest challenge for the physical AI industry; the world model helps by predicting such environmental changes and enabling appropriate responses. The performance of physical AI applied to robots or factories depends on how much corporations can narrow the gap between virtual simulations and the physical environments of reality going forward.
Han Sang-yeol, a researcher at the Software Policy Research Institute, said, "Given Korea's strengths, including the world's highest-level robot density and a fast-growing defense industry, it is necessary to secure a global competitive edge in the physical AI market by building a world model tailored to these strengths."