Photo: Lietzmann
At the completely revamped logistics congress in Berlin (if "supply chain" weren't a generic term, the "CX" suffix could also indicate its proximity to Switzerland), a science prize has been awarded for years, alongside a logistics prize. This year's prize was for AI-supported reinforcement learning.
Following presentations by the four finalists on the "Deep Dive Stage" of the BVL Supply Chain CX, the jury selected Sebastian Lang's dissertation on artificial intelligence, focusing on a specific practical application—namely, the calculation of production schedules—as the winning project. Lang utilizes methods of so-called "reinforcement learning" (RL). This allows software to be trained through trial and error so that it can subsequently calculate production process decisions in real time.
The difference from established methods is that training is not based on training labels, but rather on trial and error. This sounds somewhat like the familiar programming steps of "if-then-go-to." However, applications of reinforcement learning are intended to draw the right conclusions from both positive and negative feedback in the long term and with a significantly higher probability, developing a much more reliable planning and control strategy step by step than would be possible with conventional systems. Lang thus broke new ground: In the last 30 years, there have been only around 100 publications on reinforcement learning. Reinforcement learning is particularly well-suited for the highly demanding process planning in highly volatile, complex, or disruption-prone production environments. Lang's dissertation demonstrates, using a concrete example from transport planning, that the concept should also be transferable to other areas of logistics.
