From Passive Alignment to Active Participation: Robots Tracking Human Goal Changes for Collaborative Task Strategy Exploration

According to D. Norman's seven-stage model (1988), individuals use strategies, comprising goals and intentions within their overarching goal, to determine their next action; such strategies, moreover, are themselves adjusted by individuals based on the outcomes of prior actions. In human-robot collaboration scenarios, robots should thus follow the changing goals of humans and help them to explore their optimal collaboration strategies. However, previous research has focused on approaches that align with human intentions, neglecting the variability of the underlying goals. As a result, robots cannot adapt to human goals as these goals change implicitly. In this study, we propose a hierarchical structure to infer the actions humans can take toward a certain goal, enabling robots to support humans even when they implicitly change their goal. The proposed system estimates the most likely human goal based on the individual's current actions. When the inferred human goal changes, the robot confirms the goal with the human through conversation and then makes a proposal that aligns with the human's goals. In a human-robot table tennis scenario, this approach enhanced responsiveness to changing goals, fostering more perceptually collaborative and efficient interactions.