A box of screws, supplied by the manufacturer as bulk material, used directly for production, efficiently sorted with the appropriate software support and removed by a robot? That was wishful thinking for a long time. The task "One long screw, please!" can be completed immediately by a human looking into the crate. An industrial robot, on the other hand, cannot do this without first being explicitly taught how to do it. To do this, the process is transferred from the human to the robot - with appropriate adaptations. The robot must be equipped with additional hardware and software to replace its hands and eyes. To do this, the process is first broken down into its basic components:

• Seeing: Humans see with their eyes, which in the case of a robot, especially an industrial robot, are replaced by one or more cameras with lighting technology. The camera technology can be attached to the robot wrist, to a moving gantry application or statically above the component carrier with the objects to be removed. The cameras then record several images, which are further processed by appropriate software.
• Recognition: The desired object is identified by a three-dimensional reference model that was previously created by a corresponding computer program. Both different objects and only objects of the same type can be identified in these images. If no object is recognized in this process, feedback is sent to the corresponding control system.
• Select: Analogous to the stored reference models of the object, various gripping options are also stored. These define how the robot can pick up the object with its gripping tool. The previously identified objects are analyzed and the object that can be picked up most easily by the robot is selected. The object is selected and evaluated according to a previously defined priority list. There are areas that are stored as interference contours - for example, environments in which the robot may collide with the object removal box, steel construction elements, fences or neighboring robots. This is also taken into account when selecting the object in order to avoid a collision during object removal. If no object can be removed without collision, this information is passed on to the corresponding control system.
• Planning: After selecting the object to be gripped, a so-called path planning is carried out for the robot. For this path planning, appropriate software uses a three-dimensional simulation of the local conditions and their interfering contours. Together with the stored gripping options, the complete motion sequence is created.
• Execute: Once all previous steps have been successfully completed, the collision-free movement is executed. The robot moves to its target position in accordance with the planned movement, picks up the object and places it at a previously defined storage location in a precisely defined position. The process then starts again.

Bin picking through the ages

Screws as bulk material - picking out a single screw is a challenge for robots.

© EDAG Group

Bin picking applications have been increasingly developed over the last few years. Compared to the early days of bin picking, target recognition is now much more accurate, faster and more effective thanks to better camera technology and greater computing power. Possible collisions between the robot and its surroundings can also be calculated better and faster thanks to more powerful processors.
Long set-up times (tool changes) for gripping tools are no longer necessary thanks to flexible gripping devices and better simulation in advance. In contrast to previous automated solutions for separating, sorting and packing, bin picking is more flexible when it comes to object changes. With classic vibrating tables with separation technology, almost every change in the shape of the object requires a new tool or an adjustment. With the bin picking process, on the other hand, the new object shape and its gripping options are taught into the system and high investments for new tools are not required.