Against the backdrop of the open-source of embodied AI - specific video models, robot vision technology has ushered in new development opportunities. WeLinkirt's DaoAI 3D robot vision demonstrates strong capabilities in the unordered bin picking scenario in the automotive/parts industry.
User scenario: A leading automotive parts supplier needs to perform picking and feeding operations on various automotive parts of different shapes and sizes in the feeding process of its production line. These parts are placed randomly in bins, and the inspection objects include various key parts such as engine blocks and transmission gears.
Pain points: The open-source of embodied AI - specific video models has brought new directions for the development of robot vision technology, but the supplier still faces many difficulties. In the traditional vision system for unordered bin picking, the missed detection rate is as high as 5%, resulting in frequent manual re-inspections and increasing labor costs. The false alarm rate reaches 10%, which affects the production rhythm and reduces production efficiency. Moreover, when changing different types of parts, the change-over time is as long as 30 minutes, seriously affecting the flexibility and response speed of the production line.
Technical Principle
DaoAI 3D robot vision uses a self-developed 3D camera for imaging. The camera uses the principle of structured light, projecting a specific light pattern onto the object surface and then obtaining the three-dimensional information of the object according to the deformation of the reflected light. In terms of 6D pose estimation, a deep-learning algorithm is used to train a large number of part samples, enabling the model to learn the features of parts from different perspectives and poses. This method is effective because the deep-learning model has strong feature extraction and generalization capabilities, which can accurately identify the 6D pose of parts in a disordered state and provide accurate position and pose information for subsequent picking operations.
- Structured light imaging: Project a light pattern to obtain the three-dimensional shape of the object, providing basic data for subsequent processing.
- Deep - learning algorithm: Train a large number of samples to learn part features and improve the accuracy of 6D pose estimation.
- Brain - eye-body closed-loop: The vision system closely cooperates with the robot control system to achieve efficient picking operations.
- Sub - millimeter hand-eye coordination: Ensure high-precision robot picking and reduce errors.
WeLinkirt Solution and Product
Centered on DaoAI 3D robot vision, this solution has the capabilities of unordered bin picking, 6D pose estimation, gluing/assembly/loading and unloading guidance. In the implementation process, first, the self-developed 3D camera is used to scan the parts in the bin to quickly obtain their 3D information. Then, a deep-learning algorithm is used for 6D pose estimation to determine the exact position and pose of the parts. Next, through the brain-eye - body closed-loop technology, the vision information is transmitted to the robot control system to guide the robotic arm for accurate picking and feeding operations. At the same time, the supporting DaoAI AI AOI software system can perform quality inspections on the picked parts to ensure product quality.
DaoAI 3D robot vision provides a perfect answer to the problem of unordered bin picking in the automotive parts industry with its advanced technology and efficient solutions.
Quantitative results: After using DaoAI 3D robot vision, the detection rate of parts has increased to 99.2%, and the missed detection rate has been reduced to <0.8%, greatly reducing the workload of manual re-inspections. The false alarm rate has been reduced by -60%, effectively improving the production rhythm. The change-over time has been shortened from the original 30 minutes to 5min, significantly improving the flexibility and response speed of the production line.