Browsing by Author "Sprott, Sascha"

Now showing 1 - 2 of 2

Open Access
Observation of human behaviour in reference to human-like motions of animated avatars
(2016) Sprott, Sascha
There exist many different opinions and miscellaneous researches regarding to Moris theory about the “Uncanny Valley” - meaning the effect, when human likeness and eeriness stand in a non-linear relationship for the appearance of robots and virtual faces. Some of them depict that motion do not affect the valley in any way, but they implied it wrongly for interactions. In this thesis we wanted to investigate in the change of behaviour, when humans interact with virtual 3D avatars. Therefore we recorded various animation of a real human and mapped them onto virtual faces. Then we designed an experiment with 18 participants, who were shown four different inanimated computer generated faces and afterwards the same faces animated for an interaction, in a random order. The interaction was made with a game called “WHO AM I?”, where they slip into the role of a well known character and than had to figure out through “yes” or “no” questions who they are. After each inanimated face as well as after each interaction they had to fulfil a questionnaire, to rate human likeness, eeriness and attractiveness of the virtual avatars. We evaluated significant positive effects of human likeness and attractiveness due to interactions with computer generated avatars.
Open Access
Situation Prediction for Situation-Aware Workflows on Customer Order Settlements
(2019) Sprott, Sascha
These days, companies and manufacturers experience a need for faster and automated adaptions or re-configurations of their workflows and business processes. By means of context- and situation-aware systems, these desires are partially achievable, however only in a reactive manner. This situation based reactive behavior leads to expensive, non-efficient and time-consuming delays within workflows and business processes and hence asks for more research. Current state of the art in the topic of machine learning enables new approaches to investigate in, for workflow adaptions by predicting situations and allow proactive measurements for workflow adaptions and re-configurations. In this thesis an abstract concept is developed, that allows to turn reactive workflow adaptions into proactive adaptions within situation-aware workflow management systems, by predicting situations. This is achievable through computation of situation confidence scores by means of machine learning models and algorithms. Further, this concept allows using these algorithms without expert knowledge in the topic of machine learning, by hiding the implementation details from the user. The concept of predicting situation confidence is tested on a use case scenario for real orders of a manufacturing company and compares the classification approaches Support Vector Machine, Multilayer Perceptron and Random Forest for the prediction of orders. Results show only good performance for the Random Forest classifier, but also a concomitant possible applicability of the concept. More algorithms need to be tested and the tested algorithms need improvements, to fortify the applicability of the developed concept.