| Robot-Assisted Training (RAT) is a growing body of research in Human-Robot Interaction (HRI) that studies how robots can assist humans during a physical or cognitive training task. Robot-Assisted Training systems have a wide range of applications, varying from physical and/or social assistance in post-stroke rehabilitation to intervention and therapy for children with Autism Spectrum Disorders. The main goal of such systems is to provide a personalized and tailored session that matches user abilities and needs, by adjusting task-related parameters (e.g., task difficulty, robot behavior), in order to enhance the effects of the training session. Moreover, such systems need to adapt their training strategy based on user's affective and cognitive states. Considering the sequential nature of human-robot interactions, Reinforcement Learning (RL) is an appropriate machine learning paradigm for solving sequential decision making problems with the potential to develop adaptive robots that adjust their behavior based on human abilities, preferences and needs. This research is motivated by the challenges that arise when different types of users are considered for real-time personalization using Reinforcement Learning, in a Robot-Assisted Training scenario.;To this end, we present an Interactive Learning and Adaptation Framework for Personalized Robot-Assisted Training. This framework utilizes Interactive RL (IRL) methods to facilitate the adaptation of the robot to each individual, monitoring both behavioral (task performance) and physiological data (task engagement). We discuss how task engagement can be integrated to the personalization mechanism, through Learning from Feedback. Moreover, we show how Human-in-the-Loop approaches can be used to utilize human expertise using informative control interfaces, towards a safe and tailored interaction. We illustrate this framework with a Socially Assistive Robotic (SAR) system that instructs and monitors a cognitive training task and adjusts task difficulty and robot behavior, in order to provide a personalized training session.;We present our data-driven approach (data collection, data analysis, user modeling and simulation), as well as a user study to evaluate our real-time SAR-based prototype system for personalized cognitive training. We discuss the limitations and challenges of our approach, as well as possible future directions, considering the different modules of the proposed system (RL-based personalization, user modeling, EEG analysis, Human-in-the-Loop). The long-term goal of this research is to develop personalized and co-adaptive human-robot interactive systems, where both agents (human, robot) adapt and learn from each other, in order to establish an efficient interaction. |
