"Yes Dear, that Belongs into the Shelf!" - Exploratory Studies with Elderly People Who Learn to Train an Adaptive Robot Companion

Robot companions should be able to perform a variety of different tasks and to adapt to the user’s needs as well as to changing circumstances. To achieve this we can either built fully adaptive robots or adaptable and customizable robots. In this paper we present an adaptable companion which uses a decision making algorithm and user feedback to learn adequate behavior in new tasks. Using two different scenarios (household task, card game) the system was evaluated with elderly people in exploratory studies. We found that the perception and evaluation of the robot’s learning progress depends on the interaction scenario. Additionally, we discuss improvements for the algorithm in order to make the learning behavior appear more natural and humanlike.