Content area
Full text
Introduction
The ability to pick an individual item from a randomly orientated selection of similar items in a container and manipulate is a trivial task for humans but poses a complex problem for a robot. Known as random bin picking or simply bin picking, this is a highly desirable capability as it eliminates a repetitive manual task or the need to pre-arrange the components, thereby reducing costs. Furthermore, in the USA, for example, around 38% of the manufacturing workforce is involved in moving parts between bins and machines and with labour shortages there and in many other industrialised nations, the ability to automate this process can only grow in importance. A recent study values the global robotic bin picking market in 2023 at US$1.7bn which is forecasted to grow at a CAGR of 12.5% to reach US$5.6bn by 2033.
Bin picking systems must be able to identify and locate components in an infinite number of orientations and reach into all parts of a bin and grasp them while avoiding collisions with the bin, other components or the work cell. Furthermore, components can come in a near infinite range of materials, shapes, weights and sizes with differing physical and optical properties, compounding further both the location and grasping processes and the ability to develop general purpose systems. Despite many commercial developments enabled by advances in imaging, signal processing and gripping technologies, together with an extensive academic effort, this problem remains only partly resolved, and further work is required before bin picking becomes a widespread and routine application. This article aims to provide an insight into the state of bin picking technology by considering recent academic research and corporate developments.
Recent research activities
As with many other fields of robotics research, artificial intelligence (AI) techniques are being applied to bin picking and feature strongly in the recent literature. Workers from the National Taiwan University of Science and Technology have reported work which addressed the difficulties of picking planar objects which lack more complex geometrical features (Le and Lin, 2019). A depth sensor was used to retrieve 2D information used for image processing and to extract 3D information to perform the picking tasks at the final stage of the cycle. The 2D information was processed by a deep...