Picture of Dr. Nico Piatkowski, Chair for Artificial Intelligence at the TU Dortmund

“Perpetuate the flow of material”

Real time in logistics: The benefits that can be gained, especially in warehouse logistics, as explained by Dr. Nico Piatkowski, chair of artificial intelligence at TU Dortmund University.

What does real time mean in logistics?

Real time is basically a relative term. Just technically speaking, real time means the time a system needs to initiate an action within a specified period. This is in the nanosecond range. In an application, of course, I always need to consider the context and the specific requirements. A driverless delivery vehicle or a drone probably needs to make 60 to 100 decisions per second about parameters such as direction and speed – and each of these decisions needs to be correct. When I think globally, such as a container ship sailing around the world, I have days to optimize the routes used to forward goods once they arrive at the port of destination.

At the TU Dortmund, Dr. Nico Piatkowski researches machine learning (ML) for resource­ constrained systems. According to Piatkowski, AI and ML should not imitate people as such, but their consumption of resources. While Google’s AlphaGo did indeed defeat an international champion of the board game Go, it used 50 times the energy of its “colleagues,” who as generalists can solve much more than just one computer task. 

You specifically addressed warehouse logistics in a pilot project with one client. What is that about?

We optimized a warehouse together with an internationally renowned logistics service provider. The task was to perpetuate the flow of material from a high bay warehouse to workers at the packaging stations in order to keep the people there continuously busy. The critical thing was to come to a dynamic arrangement in the warehouse. Dynamic means, in the basic requirement for a large grocery warehouse, for example, something like placing seasonal goods – say, chocolate Easter bunnies, to be topical – in the locations in the warehouse that can be reached the fastest. But there are also much more subtle correlations that a person may not even realize. This is why we also installed a system that learns in real time and which can favorably position goods that are in high demand at certain times in variable cycles proactively.

How are learning warehouses controlled? 

It could have been done with a central computer. Aside from an energy requirement that is high but, in the long run, not more economical, such a system also has the disadvantage of not being scalable with the size of the warehouse – for warehouses of different sizes, you would need central computers of different sizes. This is why we spread the problem out and gave each storage unit a minicomputer. These minicomputers do not need to compute any comprehensive, overall solution, only a partial solution. They then communicate it only to the storage units in their immediate vicinity. If I then have adequate natural light in the warehouse and can give the computers a solar solution, I end up with an autonomous self­learning system.

Are there special infrastructural requirements for this, such as 5G?  

5G is more important for use on the road if I want to control driverless vehicles or delivery drones. Although I also need to integrate intelligence into the respective cars in that case. On one side, even a high­performance computer routing an entire fleet of drones reaches its limit at some point. And on the other side, a 5G connection can naturally be interrupted at times. This is why I need at least enough intelligence in a delivery drone that it can safely exit traffic if there is a problem.

More Information: www.tu-dortmund.de/en

Author: Heinz-Jürgen Köhler
Photos: Nico Piatkowski

Do you visit t-systems.com outside of Germany? Visit the local website for more information and offers for your country.