Abstract: Intelligent welding is important in promoting the "industrial base strengthening" project, supporting national construction and defense security. From major equipment to fine structures, welding is an indispensable key technology. As an important carrier of intelligent welding, robots will play an important role in promoting "replacing people with machines and liberating people with machines." Starting from the key technical perspectives of scene modeling of the entire welding manufacturing process, in-situ perception of welding process shape, adaptive regulation, and process knowledge construction, this paper focuses on the research status of common technologies such as "teaching-free" programming environment perception, point cloud registration, weld trajectory planning, and weld adaptive arrangement of welding robots. Then, taking the multi-source information monitoring and control system of the intelligent welding manufacturing process as an example, a hierarchical structure model of a welding manufacturing system based on IIOT-MAS (Industrial Internet of Things-Multi-agent System) is proposed, and common scientific problems such as welding multimodal information perception, fusion, and process knowledge modeling are introduced. In addition, the paper introduces a typical case of online learning of on-site perception data of engineering machinery parts welding and a welding quality evaluation model driven by a model-data dual drive. In conclusion, we explore the current trends and obstacles in advancing intelligent robot welding.