Gaming companies are entering the race to create robot brains, as their expertise in real-time simulation of virtual world physics is proving directly applicable to building virtual environments for training robot behaviors.
Industry sources reported on February 22 that NC AI, NCSoft’s AI-focused subsidiary, has officially joined the competition to develop world models, a key technology in Physical AI. NC AI recently formed the K-Physical AI Alliance consortium to participate in the “Development of World Foundation Model Technology for Physical AI Model Learning” project, overseen by the Institute of Information & Communications Technology Planning & Evaluation (IITP).
NCSoft’s entry into World Model development is backed by its game physics engine technology, which has enabled it to run virtual worlds with hundreds of thousands of simultaneous users in real-time. The reinforcement learning experience gained from massively multiplayer online role-playing games (MMORPGs) environments is particularly relevant to teaching robots optimal actions in complex real-world scenarios. Both processes rely on improving behavior through trial and error in virtual environments.
The shift in Physical AI competition from algorithms to securing physical data also opens doors for game companies. Unlike text-based generative AI, the behavioral data needed for robots to perform real-world tasks is challenging and expensive to obtain. In industrial settings, repeated experiments for robots to learn new tasks can risk equipment damage and reduced productivity.
Virtual spaces, however, can repeatedly generate diverse task scenarios, including disaster situations and irregular process conditions that are difficult to replicate in real life, enabling robots to amass crucial learning data before real-world deployment.
NCSoft is leveraging its game physics engine-based simulation technology to create a robot intelligence learning platform for manufacturing and logistics applications. Its strategy aims to enhance robot task performance by utilizing data from domestic manufacturing sites, including semiconductor and automotive industries, for AI training.
The K-Physical AI Alliance consortium formed by NC AI includes robotics and simulation firms such as Function Bay and Rainbow Robotics, alongside companies such as Samsung SDS, POSCO DX, and Lotte Innovate. These companies plan to conduct technology demonstrations in factory and logistics settings in the near future.
