en.Wedoany.com Reported - Physical AI, defined as "technology that enables AI to recognize, understand the physical world, and make autonomous judgments and actions," is attracting increasing attention. Against this backdrop, software development company Asteria and GPU cloud service provider HiRezo jointly held a media seminar on June 17, 2026, focusing on building physical AI infrastructure. Nobuhiro Kawamoto, Head of Product and Technology Strategy at HiRezo Alliance Headquarters, pointed out at the seminar that data used for training AI is the core of corporate competitiveness, emphasizing that "such data cannot be entrusted to foreign cloud services," thereby highlighting the urgency of advancing domestic computing infrastructure.

Kawamoto outlined the evolution of intelligence in robots handling real-world problems. The first generation relied entirely on manually written programs to describe actions, suitable for standardized environments like factories, but unable to handle unpredictable scenarios such as homes or roads. The second generation, starting around 2012, leveraged deep learning and reinforcement learning for autonomous learning, with machines learning recognition through vast amounts of image data and improving capabilities through trial and error. However, since they did not fundamentally understand the world, they still required enormous numbers of attempts.

Kawamoto noted that the current third generation, by introducing Transformer technology, has formed VLA (Vision-Language-Action) models that integrate vision, language, and action. Machines can now understand language commands like "pick up the cup" and have begun leveraging common knowledge from the internet to handle objects encountered for the first time.

This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com