Adapting explanations’ level of detail in a longitudinal in-the-wild office delivery robot: Ongoing results

Ferran Gebelli, Anais Garrell, Séverin Lemaignan, Raquel Ros

Keywords: Explainable Human-Robot Interaction, Longitudinal in-the-wild studies

TL;DR: This longitudinal study evaluates an office delivery robot that personalises explanation detail based on user knowledge, finding that personalization provides a more adequate verbosity while maintaing understanding.

Abstract: We present an ongoing longitudinal in-the-wild study of an office delivery robot that adapts the level of detail of its explanations to users of failure or unexpected events. We compare three explanation strategies: minimal “what happened” explanations, fully detailed including “what + why” reasons, and a personalised variant that tailors detail based on tracked user knowledge. Additionally, some users can request on‑demand extra follow-up explanations. Ongoing results from the first two weeks indicate that personalised explanations preserve both subjective and objective understanding while providing a level of detail closer to correct when compared to fully detailed explanations. Moreover, users who receive personalised explanations request fewer extra follow-up explanations compared to the group receiving minimal explanations. Full study results will confirm these results and provide their evolution through the next 2 weeks of deployment.

Submission Number: 4