"The automated handling of standardized packaged goods, such as boxes, pallets or crates, has long been common practice in intralogistics processes.
Machine learning now enables robots to master the supreme discipline: the reliable recognition, gripping and packing of individual items without any prior knowledge and quickly like humans.
Spring of this year is an exciting time for everyone active in the context of intralogistics. Two large events showed the latest developments and trends relating to the automated handling of objects and what is now possible here.
On the one hand, there was the Logimat trade fair in March in Stuttgart, which had a record number of visitors with more than 67,000 guests. In particular, it was a demonstration of mobile robots that are mainly used to transport goods, but also act as cleaning robots, for example. On the other hand, the competition for the Ifoy Award took place in Dortmund in April. This is, so to speak, the “Oscar” of intralogistics, which enjoys great attention.
A jury with representatives from 19 countries chooses the most promising innovation, for example related to forklifts.
In line with technological developments, autonomous mobile robots (AMR) and so-called order-picking robots, i.e. robots that can grasp objects, have recently also entered the international competition.
The test area on which the mobile systems have to prove themselves is now over 10,000 square meters in size. The winners will be announced in June.
44 percent market growth for logistics robots worldwide
AMR have actually been the largest growth market for service robotics, i.e. robots outside of the classic production environment, for years. The “International Federation of Robotics” counted a total of around 86,000 robots sold internationally in the logistics application field in 2022, an increase of 44 percent compared to the previous year. The majority of this is accounted for by AMRs that operate in areas without public traffic, namely around 47,500 units (an increase of 28 percent compared to 2021). In a market analysis that IFR prepared together with Fraunhofer IPA, around 400 AMR manufacturers are listed, out of a total of around 1,000 service robotics manufacturers.
But what is driving this level of automation in intralogistics upwards? Significant influences are likely to be largely known: moving objects such as goods or components is a key component in production and supply chains. The labor shortage, which is also very noticeable here, is countered by a constantly growing e-commerce, which is also accompanied by cost and time pressure.
More automation would therefore only be logical and has been implemented for many years. For example, operating devices for high racks or automatic small parts warehouses (autostore systems) achieve a reliability of 99.999 percent. This corresponds to approximately one manual intervention for every 100,000 automatic handles.
Depending on the industry and wage structure, such an extremely high level of reliability is expected from automation solutions. AI-based picking robots do not yet achieve this level of reliability today, as gripping a wide variety of items is extremely demanding for AI in object recognition and also in gripping technology. If the goods are in standardized packaging such as boxes, crates or containers on pallets, they can usually be easily handled by robots and brought to the truck.
Handling a wide variety of products without prior knowledge thanks to AI
Now one might ask why the further development of robots for intralogistics is needed and in particular what role AI could play here. The answer is strikingly simple: So far, the use of robots ends where the individual product begins. In order to recognize and grasp individual objects, robots need sufficient prior knowledge of the shape, size, color or texture. Providing this knowledge is still feasible for a handful of products. However, it is absolutely uneconomical to generate such knowledge for the tens of thousands of different products that are in warehouses and that change more or less frequently. There are technologies for this, such as learning stations, which automatically record all the necessary information about a product. But even that is still too much effort and the costs and benefits do not match.
And this is exactly where you can see why AI-based robotics is a game changer, especially in intralogistics (and therefore currently far more than for robots in production). Because it makes all previously required knowledge superfluous. So AI closes gaps in the end-to-end process by it enables the robot to handle contents of the standard packaging mentioned.
You could see how well this now works at the Logimat trade fair mentioned above. The Fraunhofer IPA demonstrator for robot-based packaging ("bin packing") was able to quickly and reliably pack never-before-seen objects in an orderly manner - 1,300 items per hour, i.e. with a cycle time of less than three seconds per handle. A person can do this over a short period of time, but not for several hours or even in three shifts.
Object recognition and grasp planning
Two technologies are particularly crucial for this individual and flexible handling of individual products. On the one hand, these are machine learning processes that enable products to be recognized without prior knowledge. This recognition has actually already reached a human-like level. This is primarily due to the huge amount of image data on the Internet, which can be used as a basis for training neural networks.
On the other hand, the robots need gripping skills - so they have to recognize where good gripping points are and plan their movements accordingly. This still needs to be expanded compared to object recognition.
There are two main reasons for this. Firstly, the products are not designed to be easy to grasp. E-commerce in particular is about consumer products that are primarily intended to look good or have direct customer benefits, but are usually not designed to be particularly “automation-friendly”. In addition, for cost and environmental reasons, the products are packaged as light and cheaply as possible, which makes it much more difficult for a robot to grasp them. For example, containers are typically sucked in from above, but this doesn't work when they are open, such as with dairy products. A so-called rolling gripper can be used for this, which picks up the container from below. And secondly, there simply isn't a perfect gripper like we humans have with our hands. This is so far unsurpassed when it comes to versatility. There is also a lack of data about how a specific item should be handled and which could be used to train neural networks.
Automatic loading and unloading of trucks is still a white spot
And finally, one last additional technology field is increasingly coming into focus: It was stated above that robot-based handling currently stops at the transporter. Here too, there are initial application developments that are testing robot-based loading and unloading of transporters.
While autonomous forklifts within logistics halls are state of the art, driving autonomously into a truck swap body is anything but trivial.
In this respect, intralogistics is and remains the most important hotspot for AI-based robotics, especially since, in addition to the above-mentioned object recognition and grip planning, it can also optimize AMR route planning or support processes as well as order planning or the detection of anomalies." [1]
1. KI-basierte Robotik: Intralogistik profitiert am stärksten. Frankfurter Allgemeine Zeitung (online) Frankfurter Allgemeine Zeitung GmbH. Apr 24, 2024. Von Werner Kraus
Komentarų nėra:
Rašyti komentarą