Order fulfillment has always been one of the core activities of the entire supply chain operations and is becoming more and more intensive thanks to the proliferation of SKUs, order volume and operational density. A typical pharma company that deals with thousands of specific, small but high volume of SKUs demand an extremely high precision order fulfillment operation. The manual process for order fulfillment here not only reduces throughput but also leads to process inefficiencies and inaccuracies. Adopting robotic bin picking solutions that deliver robotic picking of objects with the aid of computer-based vision systems and sensors delivers the right solution for picking objects that are dumped in an unstructured way.
Robotic bin picking is extremely useful while dealing with heavy, sharp, hazardous materials and to replace labor-intensive order fulfillment through picking, and bulk parts sortation. It ensures high levels of picking accuracy, throughput, high uptime with MTBF of up to 75000 hours.
Despite the best of its benefits, bin picking is still at nascent stages and is yet to realize full potential due to the sheer complexities involved.
Localization: A bin-picking robot has to identify the position and orientation of the object placed in the bin. This is an extremely unstructured environment where the positions and orientations of the objects keep on changing every time an object is picked from the bin by the robot. This requires training the system with thousands of orientations for each object or SKU, and, in an industry like E-commerce, where the business deals with lakhs of SKUs, it’s a daunting task.
System Integration: Robotic picking solution demands a good balance to coordinate the functionalities of vision systems, software, computing power and data crunching all in real-time. Ultimately the performance is delivered to grip the objects from a bin.
In fact, so far, a good amount of success rate has been achieved on geometrically symmetrical objects, which are with plain features, are not too heavy, and have some sort of sufficient planar surface in all of their random orientations, that makes it easy for the robots to pick and easy to grip feature.
There are a variety of subsets of bin picking systems that are already existing:
Structured Bin Picking: Where the robotic bin picking happens on structurally organized objects which are easy to identify and pick. Using 2D Vision, imaging, and analysis it can be done to a fair degree of success.
Semi-Structured Bin Picking: Where the objects are positioned with a fair degree of organization and predictability to aid in picking
Random Bin Picking: Where the objects are placed in completely random positions, can be overlapping, and have multiple orientations, making this the most complicated version. Advanced technologies consisting of 3D imaging and 3D analysis will have to be created to tackle the most challenging parts yet – the shingled, packaged, or deformable parts – that are difficult to capture with machine vision. This system is called Randomized Bin Picking because the robot doesn’t know what object to pick, it will pick the objects randomly from the bin based on an algorithm that provides information of position coordinates of the object to be picked.
The way forward: Bin picking normally is referred to as ‘random bin picking’ and is the holy grail of the entire research on bin picking. With the rise of the e-commerce industry, where millions of SKUs and orders need to be processed, the most challenging part of the system is that the gripper of the robot should be able to hold a square shape soap and a conical structure of a liquid container. The same applies with other industries where bin picking can yield higher efficiencies.