Compared with traditional robot solutions, the traditional robots mostly use teaching solutions, and the engineers teach them point by point. The robot completes the mechanical repetitive actions, and the application is limited.
The intelligent robot solution based on robot vision combines the camera hardware and algorithm software to give the robot vision, allowing the robot to think like a human and complete complex actions, which greatly broadens the application scenarios of the robot and meets the needs of enterprise flexible manufacturing.
The system’s visual recognition algorithm can complete the 3D position recognition of chaotically placed or closely arranged SKUs without any prior input, and effectively avoid the impact of cable ties, packaging tapes, tapes, etc. on the visual system, achieving 99.99% Success rate of visual recognition. At the same time, the system vision can achieve full coverage of 1.8m high standard pallets, which can meet the needs of more than 90% of middleware warehouses for destacking.
Product innovation and technical level overview:
Algorithm advancedness: no need for pre-stage template collection and size entry process, which is different from other domestic visual companies’ template comparison and size comparison schemes, which can effectively improve the flexibility of equipment, increase the speed of equipment deployment and debugging.
Hardware advancement: The company’s self-developed 3D large field of view camera system can adapt to the stack type up to 1.8 meters in the warehouse.
The impact of the project implementation on the industry and the driving role:
1). It has realized the recognition of massive SKUs, is insensitive to the light intensity in the warehouse, and the surface color, packaging tape, tape and other foreign objects on the surface of the box have no effect on the visual system, so that the robotic arm Mixed depalletization really has the possibility of promotion in the warehouse;
2). No prior data input, does not affect any operation process in the warehouse, solves the problems of worker handling efficiency and labor intensity, and provides industry standards for the selection of middleware warehouses It also provides a reference for the large-scale application of robot 3D vision in other category warehouses;
3). The large-field camera combined with the machine vision algorithm provides a reference for machine vision in terms of large-field and high-precision recognition.
Project technical route and innovation points:
1) The industry’s first visual recognition algorithm without prior input: 3D visual recognition technology is currently widely used in the loading and unloading of workpieces, and the robot arm automatically recognizes and grabs messyly placed workpieces, which usually requires a three-dimensional digital simulation of the workpiece Introduce the vision system, and then restore the shape of the workpiece through the features of visual recognition, so as to complete the grasping of the workpiece.
For the logistics industry, due to the huge number of SKUs, it is impossible to complete the three-dimensional digital model of the component. Therefore, in terms of visual recognition for the logistics industry, most manufacturers’ strategies are currently based on template comparison, through the early collection of each SKU surface, The data is stored in the database, and the features recognized by the camera are compared with the template during work to restore the SKU posture to obtain coordinates.
Deep learning is used to train the visual algorithm, but the same problems encountered are the time and economic cost of massive SKU template entry In addition, after the entry is completed, the SKU in the warehouse will change in real time or the irregular and unpredictable outer packaging will affect the reliability of visual recognition, so it is not suitable for in-warehouse applications. The visual recognition algorithm without a priori input circumvents the above problems very well, so that no matter how the number and type of SKUs in the warehouse change, it will not affect the visual recognition, making large-scale applications possible.
2) Large-field scene depalletizing uses standard 1.2*1m pallets in most warehouses, and the pallet height can reach about 1.8m, so the visual recognition range must cover 1.2*1*1.8m cubes. The reason is that in the large-field scene, the sensitivity to light and foreign objects is getting higher and higher, and it is more and more susceptible to interference.
Therefore, the visual system developed by our company focuses on the advanced and advanced algorithms in the large-field scene. In terms of stability, the final result is a 99.99% success rate in the commercial situation of a large-field scene.
