To main content

Auto3D: Random bin picking

As manual labor is getting increasingly expensive, the use of robots for automatic handling of parts is becoming crucial for the piece manufacturing industry in high-cost countries. One challenge that remains to be solved is to automatically pick and place objects that are placed completely unorganized in a container. This problem is commonly referred to as random bin picking.

 As a part of the Auto3D project, SINTEF and Tordivel AS have developed the 3DMaMa algorithm which is a rapid, versatile and robust method for 3D object localization and pose estimation. The input to the algorithm is a CAD model of the object to search for, along with a 3D image of e.g. a container full of parts. To capture the 3D images, we usually use structured light, but other methods for 3D acquisition are applicable as well (e.g. laser triangulation).

The algorithm works by searching through the scene for pairs of points and surface normals that have a given distance and orientation with respect to each other. Whenever such points are detected, a candidate pose is generated for evaluation against the CAD model of the object.

The 3DMaMa algorithm has been implemented in version 8 of Scorpion Vision Software, a commercial machine vision software package made by Tordivel AS.

Input: CAD model

Input to the algorithm: CAD model with two pairs of points and surface normals selected by the user.

Randomly oriented car parts

Randomly oriented car parts from Kongsberg Automotive


Result of 3DMaMa algorithm
3D image with localized car parts annotated in different colors.

The figures above show the example of the algorithm applied for Random Bin Picking, where a series of randomly oriented automotive car parts are located with 6 degrees of freedom in less than one second. The CAD model of the car part is annotated on the 3D image in different colors in each position and pose that has been identified. Coordinates and euler angles can be sent to a robot with a suitable tool for gripping the components.

Key Factors

Project duration

01/01/2010 - 01/12/2010

Explore research areas