Configuration of Adaptive Interaction by Means of Deep Learning
Diploma Thesis by Tim Gröger submitted in December 2022.
While society continues to age, more and more caregivers are needed to take care of the elderly population. However, in the future, there will not be enough caregivers to accomplish this task. A remedy can be social assistance robots, which can support elderly people in the activities of daily living. Because of users can have different constraints and properties, the social assistance robot needs to know which interaction can be used to establish optimal human robot communication.
The aim of this work is to develop a concept of how a suitable interaction can be selected for a user with certain constraints and properties using deep learning. A feature model was developed that shows valid configurations for verbal interactions with elderly people. Furthermore, several deep learning models were compared and a neuronal collaborative filtering model was chosen. This model was extended to include, in addition to collaborative filtering, the features of an interaction and the users’ properties and constraints when evaluating an interaction. To test and train this model, data was generated using studies and statistics.
The model requirements have to have a defined concept for the testing phase.