NAO robot gaze study: Humans seek confirmation looks during word games

A new study from Jens V. Rüppel and colleagues at Örebro University investigates how a robot's gaze influences human visual attention during a demanding collaborative task. Participants played a word association game with a NAO robot that served as an embodied, LLM-driven conversational partner. The experiment compared two conditions: mutual gaze (robot looks back at the human) and referential gaze (robot looks at the word cards it suggests). The team recorded participants' gaze using Tobii Pro eye-tracking glasses and analyzed fixations, speech segments, and key events.

Contrary to expectations, the robot's gaze orientation did not affect the time it took participants to first fixate on the words the robot proposed. However, the researchers found a strong behavioral pattern: participants gazed significantly more often at the robot when their own speech contained confirmation requests (e.g., "Is it that one?") compared to when it did not. This suggests that humans instinctively seek social validation from the robot in uncertain moments, even when the robot's gaze is not providing spatial cues. The authors note that the cognitively demanding nature of the word game likely made verbal coordination the dominant channel, overshadowing the more subtle effects of referential gaze.

The findings, accepted as a Late-Breaking Report at the 35th IEEE RO-MAN 2026 conference, offer practical insights for designing robot gaze behaviors in task-oriented human-robot collaboration. Rather than assuming that referential gaze alone improves efficiency, designers may need to prioritize clear verbal and mutual gaze cues to support confirmation-seeking behavior—especially in high-cognitive-load scenarios.