University of Rostock / Institute for Visual and Analytic Computing / Chair Mobile Multimedia Information Systems (Head Prof. Dr.-Ing. Thomas Kirste)
- Mr Dr. rer. nat. Sebastian Bader
- Mr Syed Ali Zafar, MSc.
Tasks and research content:
In addition to the implementation of the individual therapy measures with the help of the possibilities of humanoid robots, it is essential for a successful therapy that the robot assistant can adapt its behavior to the current situation. Therefore, the recognition of the current user and environment state is essential for a successful use of the system. For this purpose, the robot must not only recognize with whom it is currently interacting, but also the current situation in which the interaction is taking place. On one hand, the environmental state is important (location, time, people present, brightness ...), but in particular also the current training performance and the cognitive and emotional situation of the user. This information enables an analysis of the current training load and the development of the training performance as well as a corresponding feedback to the patient and an adjustment of the therapy measures. In this way, we aim to increase the user's motivation and compliance and improve the success of the therapy.
Methods of machine learning and artificial intelligence are used to interpret the inherently vague sensor data, especially from the field of Bayesian time series analysis and partially observable decision processes. These are parameterized by symbolic-logical causal models of the possible actions. The necessary techniques for the synthesis of such probabilistic filtering systems based on causal everyday models are the focus of the work in the department of Mobile Multimedia Information Systems.
Our work content includes:
- The development of sensor models for situation awareness and performance analysis in therapy exercises.
- The adaptation of methods of partially observable decision processes to the requirements of therapy support in E-BRAiN.
- The analysis of methods of (inverse) reinforcement learning for the adaptation of feedback and adaptation strategies to experiences from therapy sessions.
- The integration of decision models from psychology motivation research and preference research into algorithmic decision models for feedback selection and adaptation.