Can You See How I Learn? Human Observers' Inferences About Robot Learning Processes

Abstract

Interactive robot learners often exhibit learning behaviors that are not intuitively interpretable by human observers, which can result in suboptimal feedback in collaborative teaching settings. Yet, how humans perceive and interpret robot learning behavior is largely unknown. In a bottom-up approach with two experiments, this work provides a data-driven overview of the factors influencing human observers' understanding of the robot learning process. A novel, observation-based paradigm to directly assess human inferences about robot learning was developed. In an exploratory interview study $(N=9)$, four common themes in human inferences were identified: Agent Goals, Knowledge, Decision Making, and Learning Mechanisms. A second confirmatory study $(N=34)$ applied an expanded version of the paradigm across two tasks (navigation/manipulation) and two Reinforcement Learning (RL) algorithms (tabular/function approximation). Analyses of 816 responses confirmed the reliability of the paradigm and refined the thematic framework, revealing how these themes evolve over time and interrelate. Our findings provide a humancentered understanding of how people make sense of robot learning, offering actionable insights for designing interpretable RL systems and improving transparency in Human-Robot Interaction (HRI).