Industrial agents: Architecture, key technologies, and future perspectives

The advancement of new-generation artificial intelligence technologies, such as industrial foundation models and embodied AI, has catalyzed the emergence of new agents oriented towards Agentic AI, presenting fresh opportunities for the development of industrial artificial intelligence. As novel agent systems driven by these cutting-edge technologies and serving the full lifecycle of industrial products, industrial agents encompass both embodied and non-embodied (software system) forms. They possess core capabilities tailored to industrial scenarios, including perception and understanding, knowledge memory, planning and decision-making, communication and interaction, autonomous execution, and evolutionary learning. Consequently, they provide intelligent services across the entire business chain of industrial manufacturing, including R&D, manufacturing, testing, management, and maintenance. Currently, systematic research regarding new-generation industrial agents remains a gap in the field. This paper defines the concept of industrial agents and analyzes their key characteristics. It proposes an architecture for industrial agents comprising three layers: the industrial agent base, the industrial agent system, and industrial applications, while also elucidating the associated key technologies. Centering on industrial product lifecycle activities, the paper discusses typical future application scenarios for agents in R&D, manufacturing, testing, management, and maintenance. Finally, the paper provides an outlook on the future development of industrial agents, analyzing prospective technical directions such as the industrial agent foundation, industrial world models, software-hardware synergetic agents, trustworthy agents with long-term memory, endogenous security, and standardization. This work provides a systematic reference for the construction of theoretical and technical frameworks, system development, and industrial application of industrial agents driven by new-generation AI.