Mitsubishi Electric aims to “smartify devices and the edge, and develop AI that evolves together with humans,” and is advancing both research and practical implementation of edge AI. In recent years, the company has introduced its proprietary technology brand, Neuro-Physical AI, which emphasizes reliability and safety in the physical space, creating new value by realizing physical AI that connects the cyber and physical worlds. This lecture will cover research on manufacturing-oriented language models, high-reliability edge AI technologies for FA equipment, application to visual inspection of finished products, fault diagnosis, and studies on performance assurance for edge AI. Through these topics, we will discuss the evolution of edge AI—achieving real-time performance, reliability, and human collaboration—enabled by Neuro-Physical AI as a function of Maisart, and its development from research to on-site implementation.