Superb AI “Proprietary AI Foundation Model Project” Phase 2: Key Takeaways on Digital Twin Assetization

Superb AI is advancing the development of Physical AI by focusing on simulation data to overcome the limitations of real-world data collection, which is costly and time-consuming. Through their Proprietary AI Foundation Model Project, Superb AI has progressed from collecting high-resolution RGB-D data in Korean residential environments to the second phase, which involves converting this data into digital assets for simulation environments. This phase centers on creating three core digital assets—Space, Action, and Object—enabling the construction of a robot-trainable virtual environment, or digital twin. These assets allow for the generation of diverse scenarios that are not feasible in the real world, thereby expanding the scale and diversity of data for robot learning. The project aims to reduce the constraints of data acquisition and stimulate Korea's robotics and AI ecosystem by making high-quality datasets accessible. By leveraging techniques like 3D Gaussian Splatting and SMPL, Superb AI is addressing the Sim-to-Real Gap to ensure that simulations align closely with real-world conditions. The ultimate goal is to enable robots to learn through generated situations rather than physical filming, positioning Korea as a leader in Physical AI development.