Tasks that are easy and simple for humans to perform are often the most difficult to automate. For instance, it is easy for the human eye to detect a slip sheet of carton or plastic hanging down from a layer on a pallet, but very difficult for a machine. This problem is now solved with artificial intelligence (AI).
When our Layer Picker gets hold of a layer with a slip sheet hanging down, it automatically stops as it can’t identify it as goods and thus identifies it as a risk of dropping a layer. On average, an operator of the Körber Layer Picker spends two minutes reacting to the alarm and resetting the machine each time it occurs.
With a digital add-on product based on AI called the Operator Eye, the equipment can reset itself in just seconds.
Demystifying AI
Since the launch of Chat-GPT, AI has gotten a lot of hype and attention. Will it take my job, will it develop a conscience, will it take over the world? I assume some of these fearful questions are based on difficulties of understanding what is going on in the algorithms behind the scenes.
With the Operator Eye, we are bringing AI to the warehouse and there is absolutely nothing mysterious about it.
Operator Eye is based on vision technology with four cameras attached to the Layer Picker. Vision technology is not new, but without AI we would need to have reference pictures for every single way the slip sheet could be picked up by the Layer Picker – including every angle and every lighting situation. If we couldn’t provide the model with a close match, we couldn’t define an action.
Today, due to the easy accessibility of computer power and the amount of available data this has changed. Combined with the rapid increasing numbers of AI classification models, developed for everything from self-driving cars to ChatGPT that simplifies building and deploying machine learning models, we can train the algorithm. It can now recognize patterns and calculate a very accurate probability that the Layer Picker has picked up a layer successfully with hanging plastic or carton and that it is not a box of ketchup bottles about to fall down.
The AI behind the Operator Eye is not magic. It is a lot of pictures of scenarios where your machine has stopped. Learning this creates a recognizable pattern for an algorithm based on our deep knowledge about depalletizing.
Learning from the operator
To really call the Operator Eye an AI solution, it of course has to improve over time as well. And it does.
When we deploy Operator Eye at a new site, it learns from the operators and continuously becomes more and more efficient. The cameras detect the reason for a stop, and if the operator hits the reset button, the system will automatically learn to reset next time a similar incident causes a stop.
As these learnings accumulate over time, you will experience fewer and fewer situations, where the operator needs to interact with the equipment.
