Superb AI Achieves Phase 1 Milestone in the State-Run Proprietary AI Foundation Model Project (1.08M Robot Data)

Starting in 2025, the AI landscape is set to evolve from digital intelligence to Physical AI, which involves systems that can interact with the real world, as exemplified by Superb AI's work in Korea. The company is addressing the scarcity of high-quality vision data, particularly RGB-D data, crucial for robots to perform tasks like dishwashing by understanding and manipulating physical environments. Superb AI's contribution to the Korean government’s Proprietary AI Foundation Model Project includes the development of high-quality training data for Vision-Language-Action (VLA) models and the creation of datasets tailored to Korean living spaces. By using a combination of controlled studio and real-home data, the company has refined 400 million frames into 300,000 training assets using its Auto-Curate technology, which enhances AI training efficiency and reduces costs. This initiative supports the advancement of AI systems capable of reasoning about human intent and performing actions in complex, cluttered environments. Moving forward, Superb AI aims to expand its datasets to further develop robust Physical AI models, enhancing Korea's competitiveness in AI technology.