Various neural network based methods are capable of anticipating human body motions from data for a short period of time. What these methods lack are the interpretability and explainability of the network and its results. We propose to use Dynamic Mode Decomposition with delays to represent and anticipate human body motions. Exploring the influence of the number of delays on the reconstruction and prediction of various motion classes, we show that the anticipation errors in our results are comparable or even better for very short anticipation times (< 0.4 sec) to a recurrent neural network based method. We perceive our method as a first step towards the interpretability of the results by representing human body motions as linear combinations of "factors". In addition, compared to the neural network based methods large training times are not needed. Actually, our methods do not even regress to any other motions than the one to be anticipated and hence is of a generic nature.

In industrial automation, the use of robots is already standard. But there is still a lot of room for further automation. One such place where improvements can be made is in the adjustment of a production system to new and unknown products. Currently, this task includes the reprogramming of the robot and a readjustment of the image processing algorithms if sensors are involved. This takes time, effort, and a specialist, something especially small and middle-sized companies shy away from. We propose to represent a physical production line with a digital twin, using the simulated production system to generate labeled data to be used for training in a deep learning component. An artificial neural network will be trained to both recognize and localize the observed products. This allows the production line to handle both known and unknown products more flexible. The deep learning component itself is located in a cloud and can be accessed through a web service, allowing any member of the staff to initiate the training, regardless of their programming skills. In summary, our approach addresses not only further automation in manufacturing but also the use of synthesized data for deep learning.

No abstract

Um kleine Losgrößen robotergestützt wirtschaftlich zu fertigen, müssen die entsprechenden  Steuerungsprogramme aufwandsarm anpassbar sein. Eine  besondere Herausforderung ist dabei die Umstellung  optischer Sensoren zur autonomen Erkennung und Lagebestimmung von Werkstücken, weil die  eingesetzten Algorithmen meist auf bestimmte Objekt- und  Umwelteigenschaften zugeschnitten sind. Ein  Lösungsansatz besteht in der Kombination von Digitalen  Zwillingen mit Methoden der Künstlichen Intelligenz in  einem cyber-physischen System, um neue Produkte  automatisch zu identifizieren, ihre Lage zu bestimmen und die Roboterprogramme anzupassen.