Bin-picking/line-picking

 

In automation and handling technology, forwarding workpieces to different places in the production environment is a key issue. The workpieces concerned are often located in a disorderly fashion in bins or pallet cages. To advance these components automatically, they first have to be singularized by a spiral conveyor and belt with sophisticated mechanical systems that are designed to handle a specific component. This form of automated feeding is extremely inflexible as far as component modification is concerned because adapting such feeding systems is both complicated and expensive.

An alternative is to pick up the unsorted components by robot directly out of the bin or from the conveyor. To do this, an object’s position in space has to be accurately identified. For a limited amount of workpieces, algorithms from 2D image-processing already exist. However, such methods are hardly suitable for locating randomly-positioned objects.


“Bin-picking”

The difficulty with many tasks associated with bin-picking is the limited amount of space available for a robot to pick up unsorted workpieces directly from a transport container. If the geometry of a workpiece is altered for any reason, the maximum effort required should be either adaption or replacement of the end effector. It is thus essential that the position of objects is identified reliably and quickly. For this reason, object recognition software based on the analysis of 3D data has been developed at Fraunhofer IPA.


3D object recognition

Object position recognition is based on the principle of fitting geometric primitives, such as planes, cylinders, cones or tori, into point clouds. This approach was chosen because many technical workpieces have cylindrically- or conically-shaped elements. However not all of the objects requiring recognition are just cylindrical or conical in shape; much more complex workpieces with freely-formed surfaces as well as geometric features are also recognized.


Key data of the method

The 3D object position recognition software based on geometric primitives runs on a standard computer. The calculation time required to identify and locate a workpiece is around 0.5 second. The method is recognizes an object’s position precisely. This is determined by the accuracy of the 3D sensor used. The demonstrator which has been constructed has a localization accuracy of +/- 0.5 mm.

The 3D object recognition system based on geometric primitives is currently capable of handling workpieces with mostly geometric features. Work is also in progress to enable more complex workpieces with a significantly higher proportion of freely-formed surfaces to be recognized.

Our services include feasibility studies and 3D object recognition software as well as the construction of prototypes and complete systems.