2014 R&D 100 Winner
Automated assembly systems require organized, aligned parts. In most factories, unorganized parts are usually sorted and positioned by manual labor. This task must keep pace with assembly, and is tedious and stressful. Parts tools to automate “bin-picking” have previously been custom-designed for each part. Machine vision has also been introduced, but it has been limited to parts with simple shapes.
Mitsubishi Electric Corp.’s MELFA-3D Vision system for industrial robot arms completely automates this bin-picking task. A projector creates multiple slit patterns that are projected on the piled parts, which are captured with the camera. A depth map is reconstructed by using the captured images and a structured-light decoding algorithm. Industrial parts generally have no textures which creates challenges for 3-D reconstruction systems using only cameras. MELFA-3D Vision can perceive and organize industrial parts, because the slit patterns introduce artificial textures on the parts.
By comparing the part with a 3-D CAD model, the position is established. To allow the robot arm to grasp it and position it correctly, a model matching method uses simple features and aggregates evidence in a voting framework. For complex parts, the system can task one robot arm to pick the part from the bin and another to position it for assembly.
3-D vision system for industrial robot arms
Mitsubishi Electric Corp.
The MELFA-3D Vision Development Team from Mitsubishi Electric Corp.
Yukiyasu Domae, Principal Developer
Yuichi Taguchi, Principal Developer