The core client involved in the project is BEUMER Group A/S, one of the world leading companies within Baggage Handling Systems (BHSs), and as part of DIGIT team. Our industrial PhD student René Arendt Sørensen, is working on using Machine Learning technologies for improving routing in BHSs.

The client need involves the current routing schemes, which are based on shortest path algorithms and manually adjusted through an expensive and time-consuming trial and error process. Previously, the company had to use the real physical system to find better routing schemes, but lately, they acquired an emulator, allowing the developers to find the bottlenecks and errors much faster and without interrupting a running system. In our client case, the goal is to further utilize this emulator to train a Deep Neural Network to select the good routing schemes.

The approach in providing the right solution is to use a method called Deep Reinforcement Learning, which is a method within machine learning. Instead of learning from data, Reinforcement Learning relies on experiencing an environment and storing its experience in a Deep Neural Network. To achieve this, three abstraction levels are used. First, a simple graph network is used, to find out how to design the model, then the emulator containing a digital version of a real system is used to find out how to train the model on a more realistic system, and last, the model must try to control a real system to test the results.

Client Contact: Morten Granum, Beumer Group A/S, Morten.Granum@beumergroup.com.

Online reference: http://digit.au.dk/research-projects/machine-learning/