AUTHOR=Belardinelli Anna , Wang Chao , Tanneberg Daniel , Hasler Stephan , Gienger Michael 

TITLE=Train your robot in AR: insights and challenges for humans and robots in continual teaching and learning

JOURNAL=Frontiers in Robotics and AI

VOLUME=Volume 12 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/robotics-and-ai/articles/10.3389/frobt.2025.1605652

DOI=10.3389/frobt.2025.1605652

ISSN=2296-9144

ABSTRACT=Supportive robots that can be deployed in our homes will need to be understandable, operable, and teachable by non-expert users. This calls for an intuitive Human-Robot Interaction approach that is also safe and sustainable in the long term. Still, few studies have looked at interactive task learning in repeated, unscripted interactions within loosely supervised settings. In such cases the robot should incrementally learn from the user and consequentially expand its knowledge and abilities, a feature which presents the challenge of designing robots that interact and learn in real time. Here, we present a robotic system capable of continual learning from interaction, generalizing learned skills, and planning task execution based on the received training. We were interested in how interacting with such a system would impact the user experience and understanding. In an exploratory study, we assessed such dynamics with participants free to teach the robot simple tasks in Augmented Reality without supervision. Participants could access AR glasses spontaneously in a shared space and demonstrate physical skills in a virtual kitchen scene. A holographic robot gave feedback on its understanding and, after the demonstration, could ask questions to generalize the acquired task knowledge. The robot learned the semantic effects of the demonstrated actions and, upon request, could reproduce those on observed or novel objects through generalization. The results show that the users found the system engaging, understandable, and trustworthy, but with larger variance on the last two constructs. Participants who explored the scene more were able to expand the robot’s knowledge more effectively, and those who felt they understood the robot better were also more trusting toward it. No significant variation in the user experience or their teaching behavior was found across two interactions, yet the low return rate and free-form comments hint at critical lessons for interactive learning systems.