How does Viewtrix use visual perception and industrial simulation to tackle the hard challenges of ship and marine engineering manufacturing?

On October 22, the 2026 China Marine Economy Expo was held at the Shenzhen World Exhibition & Convention Center (Bao'an). In the exhibition space of ten thousand square meters, underwater robots, ocean satellites, and AI ocean large models appeared in turn, and marine technology is moving from the laboratory to real scenarios. In this technology narrative of "going to the sea," the four sets of equipment and one simulation platform brought by Hunan Viewtrix Robot Co., Ltd. (abbreviated as "Viewtrix") are more like production units moved directly from a modern shipyard: a dual-machine collaborative welding workstation, a humanoid welding robot, an intelligent residual material scanning system, an intelligent grinding workstation, and the "Kunwu" industrial simulation platform. Unlike technical solutions focused on ocean perception and offshore monitoring, Viewtrix uses visual perception and industrial simulation to enter the most difficult-to-standardize process in ship and marine engineering equipment manufacturing, bringing cutting-edge technology to the production frontline of manufacturing workshops. Shipyard challenges: why can't automation be pushed forward? Shipbuilding is a typical non-standard discrete manufacturing industry with multiple varieties and small batches. Components such as sub-assemblies, lifting lugs, and panels have complex forms. Deformation of incoming steel plates, inconsistent groove gaps, and deviations between design drawings and actual objects are the norm in workshops. Traditional industrial robots are highly dependent on manual teaching and playback. When the workpiece category changes, reprogramming and debugging are required. Coupled with practical difficulties such as tight supply of skilled welders, harsh workshop operating environments, and high costs of trial welding and rework, automated equipment has always been difficult to deploy on a large scale. The processing of non-standard components is highly dependent on the experience of senior technicians, becoming the most prominent bottleneck in the digital transformation of the shipbuilding industry. The policy direction has long been clear. In 2025, the Ministry of Industry and Information Technology and eight other departments issued the "Implementation Opinions on the Special Action for 'Artificial Intelligence + Manufacturing'," proposing to promote the application of artificial intelligence technology in the shipbuilding industry. It calls for building large models for the shipbuilding industry, exploring new models for R&D and design, and, oriented to the needs of unmanned and intelligent production of large ships and marine equipment, advancing the intelligent upgrading of key processes such as "cutting-welding-spraying-logistics." Stripping away the complicated surface, the underlying contradictions of intelligent manufacturing in shipyards have two points: cannot see clearly, and cannot afford to test. Drawings are standardized, but the processed physical objects always deform, and machines "cannot see" the real workpieces; process plans are preset, but every verification occupies the production line and working hours, and on-site trial and error is costly. Viewtrix has built a complete four-layer industrial intelligence closed loop of "perception-execution-decision-operation and maintenance," abandoning the idea of fragmented transformation and fully presenting integrated upstream and downstream intelligent manufacturing capabilities such as cutting, blanking and sorting, groove processing, flexible grinding, and weld inspection. First, let machines understand non-standard physical objects. The problem-solving approach proposed by Viewtrix is, first of all, to push "seeing" to the extreme. If traditional automation only recognizes standardized drawings, visual perception is to install "industrial eyes" on robots, directly reading the real non-standard physical objects in the workshop. After steel plate nesting and cutting, a large amount of irregular residual material is produced. The traditional manual measurement and hand-drawn ledger model is inefficient, data is easily lost, and residual material management is difficult, which is a difficult point in lean management of blanking workshops. Viewtrix has developed an intelligent residual material scanning system for the shipbuilding, steel structure, and steel industries. Relying on high-precision line-scan cameras and self-developed 3D vision algorithms, it automatically completes residual material contour scanning, dimension extraction, and DXF modeling, links equipment to realize automatic loading and unloading of residual materials and stereoscopic warehouse storage, guides secondary cutting of residual materials, and revitalizes the value of steel plate leftovers. Specifically, multiple 3D line-scan devices work together for stitching to achieve large-format high-precision visual fusion imaging, and algorithms extract the contour edges of residual materials; residual material contours are automatically scanned and DXF modeled, replacing manual surveying and ledger creation; automatic loading and unloading and residual material warehousing are supported; automatic nesting and cutting guidance are provided, linking with the cutting system to complete secondary cutting of residual materials. Seeing the leftover material clearly is only the beginning. Grinding is an essential process in the heavy industry. Grinding cutting burrs on workpiece edges and chamfering non-welded edges are requirements for small-batch, multi-variety production. The consistency and quality of grinding are difficult to guarantee and highly dependent on manual experience. Viewtrix's teaching-free flexible grinding has a built-in free-edge rule library and automatically identifies grinding contours; new workpieces can be imported from drawings to generate grinding paths, and multiple process parameters can be switched with one click, greatly improving the switching efficiency of multi-variety, small-batch production. 3D vision locates parts, accurately extracts part contours, and automatically generates robot trajectories, working with floating grinding heads; the system supports linkage with production line loading and unloading equipment and adapts to different site layouts. Dual-machine collaboration: welding with or without drawings. Welding is one of the links where the non-standard contradictions in shipbuilding are most concentrated. Tight supply of highly skilled welders, time-consuming teaching and programming, repeated debugging of multi-layer multi-pass welding processes, and incompatibility among robots of different brands... these are all real dilemmas facing the industry. Viewtrix's dual-machine collaborative welding workstation integrates self-developed 3D vision hardware, AI algorithms, and SpeedWeld intelligent welding control software, covering the complete chain from perception and decision-making to execution, enabling robots to "understand" welds themselves and make intelligent autonomous decisions. SpeedWeld intelligent welding control software is the true core of this system. With "teaching-free, high flexibility, and high precision" as its core, it integrates high-precision vision, AI algorithms, and automatic path generation technology, supports multi-robot collaborative operations, allows customization of intermittent welding, welding length, and welding sequence, and can realize automatic collection and warehousing of data throughout the welding process. With stronger adaptability and versatility, SpeedWeld has been adapted to more than 20 mainstream robot brands such as FANUC, KUKA, YASKAWA, and ABB, requiring no opening of the robots' underlying protocols and offering plug-and-play use. Relying on SpeedWeld control software, Viewtrix's dual-machine collaborative welding workstation can flexibly adapt to both drawing and no-drawing modes. It can import CAD models and also quickly generate welding paths through 3D scanning reverse modeling. The workstation supports flexible production from standard parts to large complex spatial components. In terms of capability boundaries, it can stably identify highly reflective workpieces such as 0.5mm thin plates, galvanized steel, and stainless steel. Thin plates are prone to burn-through, galvanized steel is prone to porosity, and stainless steel is prone to arc light reflection; these three categories have long been no-go zones for automated welding; and under strong arc light environments, it is also less prone to welding deviation. In terms of efficiency and traceability, the workstation is equipped with a point cloud large model, achieving truly teaching-free operation that ordinary operators can pick up easily, shortening the changeover cycle for new workpieces by 90%, while collecting welding current, voltage, and welding speed data in real time, automatically alerting on anomalies, and automatically generating weld seam traceability reports, enabling full traceability of welding quality. From boilers and ships to construction machinery and automobiles, Viewtrix has already landed in multiple core heavy industry sectors, providing targeted solutions for the unique welding challenges of each industry and delivering multi-dimensional value in efficiency, cost, quality, and flexibility. For example, ship sub-assembly welding faces the pain points of highly diverse workpiece categories, often no drawings, and high requirements for welding stability and precision. Relying on its point cloud large model solution, Viewtrix can generate a 3D model of the workpiece through one-click reverse modeling and automatically extract weld seam paths; without preset templates, manual point cloud denoising, or manual parameter tuning, it greatly improves equipment usability and production flexibility, adapting to flexible sub-assembly welding production in shipyards. Humanoid robots—taking the welding torch out of fixed workstations. But the automation of fixed workstations still faces an unavoidable wall. Traditional fixed welding equipment is constrained by tracks, robotic arms, and dedicated tooling: it cannot enter narrow, enclosed cabins, cannot reach oversized irregular components, and cross-station processing requires re-disassembly and reassembly. Viewtrix unveiled two humanoid robot products. Among them, the full-size humanoid welding robot developed for heavy industry scenarios adopts a 1.65-meter full-size bipedal humanoid configuration, with a 10 kg arm payload, full-size high flexibility, self-developed large and small brains, visual navigation and positioning, autonomous intelligent welding capabilities, and industrial-grade protection, adapting to scenarios such as ship sections, heavy steel structures, and construction machinery; in the future it will bring autonomous flexible welding capabilities to the heavy industry frontline. The wheeled humanoid robot focuses on multi-scenario services, supporting autonomous navigation, point-of-interest guidance, voice narration, fetching objects and delivering water, indoor inspection, and sorting and handling; it can be used for exhibition hall guidance, conference reception, factory logistics, and smart inspection, empowering smart office and campus services. However, humanoid robots are not black technology that appeared out of thin air. To be competent for on-site welding operations, visual perception, motion control, and welding processes must undergo long-term integration. Viewtrix's years of accumulated 3D vision technology, welding algorithms, and heavy industry process expertise are what support the humanoid robot's practical capabilities on the shop floor. The fixed workstation and the humanoid robot, one static and one dynamic, complement each other; Viewtrix hopes to extend the same ability to "understand physical objects" to different nodes of the shipbuilding process chain in the future. Let decisions run through a virtual shipyard first. However, no matter how accurately a single point is seen or how well it is welded, the production line may still be "planned poorly." If vision solves "how to do this piece," simulation must solve "how to arrange, adjust, and change this line." KUNWU is a one-stop production line design and development platform built by Viewtrix for manufacturing enterprises and smart manufacturing integrators, featuring simulation-as-development and design-as-delivery. Most industrial software on the market focuses only on a single link of production line development, while KUNWU "penetrates horizontally and vertically": horizontally, it connects the four major links of production line design, robot offline programming, virtual commissioning, and production scheduling simulation, relying on digital twins to achieve real-time synchronization between virtual and real; vertically, it carries two core capabilities—low-code generation and industrial large models—and can directly generate deployable production line control software from simulated processes. It addresses the pain points of traditional production lines such as high construction costs, difficult layout, cumbersome code changes for process modifications, and long on-site commissioning cycles, exposing problems in advance through virtual verification and reducing on-site rework; it supports low-code rapid process adjustment, compressing on-site commissioning from weeks to days. At present, KUNWU has accumulated extensive industry know-how, with more than 30 built-in industry scenario templates, more than 10 professional process packages, rich operator models, and an industrial equipment library, covering fields such as shipbuilding and offshore engineering, construction machinery, automobiles, and lithium batteries. Every robot workstation, inspection system, and intelligent production line delivered by Viewtrix runs on the KUNWU platform at its foundation; tempered by numerous real projects, it is by no means a laboratory concept product. Seeing the quality of "general-purpose" from one industry entry point. The quality of smart manufacturing solutions is never judged by booth demonstrations, but by shop-floor implementation. Currently, Viewtrix has served China State Shipbuilding Corporation, COSCO Shipping Heavy Industry, China Merchants Industry, and multiple leading private shipbuilding enterprises, completing large-scale implementation of multiple shipyard intelligent projects. These capabilities were not born only for ships. As a definer and leader of general-purpose smart manufacturing, Viewtrix summarizes its path as One Brain, Any Manufacturing—using the same "brain" to drive the intelligent upgrade of any manufacturing scenario. The company's self-developed industrial foundation large model, real industrial working-condition data, full-stack system engineering capabilities, and process knowledge accumulated in industries such as construction machinery, automobile manufacturing, and shipbuilding and offshore engineering are deeply integrated to build a truly implementable and scalable physical AI capability system. Today, Viewtrix's products are actually running on the production lines of more than 300 customers worldwide, with business reaching domestic and overseas markets and products exported to North America, Europe, Japan, and South Korea; it is currently one of the few full-stack physical AI companies in China that simultaneously possesses models, data, systems, and process knowledge. From starting in heavy industry, entering the construction machinery sector, to the shipbuilding business, and then replicating to more industries, this path itself is the way to observe the quality of its "general-purpose" nature. Walking into a real shipbuilding workshop, one will find that the intelligentization of offshore engineering equipment has long moved beyond concept demonstrations in exhibition halls. Vision equipment scans hull components of various shapes, industrial simulation pre-simulates production line processes, and robots take on heavy welding operations. The digital vision that once stayed in proposal reports is gradually becoming the day-to-day production reality of shipyards, accompanied by scan after scan and weld after weld.