Less and less people are willing to perform monotonous tasks, and the growing number of different items makes their work increasingly error-prone. How can robots help?
In previous blogs, we have written about future intelligent goods flows that are fully driven by e-commerce. Millions of parcels need to be handled in ever-shorter lead times. At the same time, they increasingly vary in appearance, because we are no longer dealing with similar-looking, large boxes that are sent to retail outlets. Rather, the goods that must be handled are individual items ordered by end customers. They include boxes, apparel orders, bottles, jiffy bags, plastic bags, padded mailers, flasks, and cylinders that differ greatly in shape, weight, size, material, texture, and color. This variety makes manual tasks such as picking boxes and conveyor belt work increasingly tedious and error-prone for humans. Therefore, it only seems logical to have robots perform them. After all, they will enhance quality and save costs. So why aren’t warehouses already crowded with robots?
Why not let robots stow and pick?
Quite frequently, inspiring videos pop up in our LinkedIn feeds, showing how Amazon and Alibaba have automated package sorting and transportation through automated storage and retrieval systems such as conveyor belts, shuttles, automated guided vehicles, or automated mobile robots or ants. However, a more careful look reveals that all items are still manually stowed onto these vehicles, or they're manually picked from storage systems for customer order singulation. This is the missing link in fully automated material handling.
Of course, companies are seeking solutions to replace repetitive manual work. Yet robots that stow and pick are still very scarce. The main hurdle: traditionally, robots can't handle the items' large variety in appearance, and they’re unable to reach the high speed required to stow large quantities of items that must be handled. Efficiently going through these process steps still requires a type of intelligence which is innate to humans and which robots have always lacked – that is, until recently.
Solution: AI-based learning algorithms for vision-guided robots
New startups have tackled the issue. Companies such as Fizyr, consisting of Delft-based engineers, have added deep, AI-based learning algorithms to vision-guided robots. As a result, the software can automatically classify unknown objects to find the best possible grasp location for picking.
The accuracy and robustness of the software is already reaching levels that were previously considered impossible in technical terms. Recent challenges organized by Amazon have shown the immense abilities of the current software to effectively handle a great variety of items. Soon, this machine-vision software can also be used to perform classifications and manipulations, both for inspection and quality control purposes.
Practical matters: costs, speed, and capacities
Recent practical examples have shown that vision-guided robots can reach a speed of currently 900 (but soon more than 7000) picks per hour. The resulting cost per pick of approximately 5 eurocents is already similar to – or, in some cases, lower than – the cost of human handling. Moreover, robots run 24/7 and don't need any coffee breaks, and in all probability, they will be able to reach higher quality levels.
Briefly put, we believe the missing link in fully automating warehouse operations will soon be covered. The technology is already available, and it will be implemented over the next 2 to 3 years. And then, the time has come to switch off all the lights in our warehouses!