Industrial AI Agent with large and small-scale models collaboration: Concept, framework, and application

Industrial AI Agent is an intelligent system that integrates industrial expertise with artificial intelligence technologies and can independently complete the closed loop of “perception, decision-making, execution, and learning” in industrial scenarios. By virtue of its capabilities of self-perception, self-learning, self-decision-making, self-execution, and self-adaptation, industrial AI Agent demonstrates great potential in transforming industrial production from passive response to active prediction, as well as realizing full-process autonomy and global collaborative optimization. To this end, this paper deeply integrates the capabilities of large models such as ChatGPT and Qwen, including global cognition, multimodal interaction and intelligent reasoning, with the core advantages of small models such as industrial software, machine learning algorithms and digital twin models in terms of professionalism, lightweight deployment and low-cost operation, and proposes a novel paradigm of Industrial AI Agent featuring “large models as the general commander and small models as the specific executors”. On this basis, a six-in-one architecture of industrial AI Agent consisting of physical space, perception space, decision space, behavior space, learning space, and application space is designed, and the collaborative operation mechanism between large models and small models is clarified. Finally, the rationality and effectiveness of the proposed architecture, models, and methods are verified through a case study of the autonomous assembly of key components of artificial satellites empowered by industrial AI Agent.