SILUX: Bin picking system using computer vision and deep learning

SILUX addressed the problem known as bin picking, which requires combining robotics and computer vision so that an automated system can identify, locate, and handle parts placed in a disordered way inside a container.

The objectives of the project were:

• Automate the picking of parts from a box and their transfer to the next point in the production line.
• Give the robotic arm the ability to decide which part to pick at any given moment.
• Enable the detection of any type of part without requiring technical knowledge on the part of the end user.

The areas around which SILUX was structured were:

• 2D detection and segmentation: using deep learning techniques, the system identified the parts present in the scene and determined which one was the most convenient to pick.
• 3D matching: in a second phase, the system calculated the position and pose coordinates of the selected part and provided them to the robotic arm.
• Automatic model generation: a tool fully automated the creation of the 2D and 3D detection models, transparently for the end user, which made it possible to adapt the system to any part.

Normagrup relied on CTIC, which contributed its specialized knowledge of computer vision, deep learning, and industrial robotics to the development of the solution, validated in a real production environment.