Explainability Analysis for Skill Execution

Explainability holds significant importance for autonomous robots deployed in human-centered situations, particularly when errors occur during execution. In the context of robot action, it is important to consider various levels and types of explainability. The social dimension of Artificial Intelligence (AI) and robotic explanations, which highlights how they affect social interaction, values, and decision-making, has received little to no attention in prior research. With a particular emphasis on item handover, we hypothesize that users prefer systems with explanations and that explanations in natural language are more appealing than heatmaps. A user study, involving participants from diverse backgrounds and levels of expertise, is conducted to evaluate different levels and preferred types of explainability. The study results support our hypotheses and offer additional valuable information for future system development.