ARBI · Adapting Robot Behavior based on Interaction

Having robots that are able to understand, communicate and coordinate with humans executing daily tasks is one of the dreams of society. Moreover, when people interact with robots, they expect the robots to behave in a certain manner, and tend to apply them social models, similarly as how they behave when meeting a new person. Therefore, besides typical perception, reasoning and motion abilities (which have been largely studied so far), the robots also need to have certain abilities to enhance their social behavior when interacting with people. Such abilities are: (i) being proactive, i.e. capability to propose tasks when detecting relevant context, instead of waiting for the human to explicitly ask for something, (ii) perform understandable actions, i.e. show intentions so the human can easily understand and be aware of the robot's actions, (iii) achieve social acceptance, i.e. take into account human preferences so the human feels comfortable with the robot's task execution, (iv) learn, i.e. not only to improve the current knowledge of the robot, but to increase it and to adapt its behavior to the different contexts it may encounter. The main goal of this project is to provide the robot with a reasoning approach to select the most appropriate task execution based on the humans’ preferences and social facts and to improve the robot's behavior as the experience with humans increases. To this end, we propose to design a decisional framework that includes a user modeling approach (stereotype based) and a machine learning technique (case-based reasoning). To evaluate the approach, experiments in simulation and using real robots will be performed.