Levels of explanation -- implementation and evaluation of what and when for different time-sensitive tasks

TL;DR

This study develops and evaluates levels of explanation (LOE) for time-sensitive human-robot interaction, demonstrating that higher LOE improves communication effectiveness and task performance in telepresence healthcare scenarios.

Contribution

The paper introduces a structured approach to LOE based on verbosity and explanation patterns, and empirically evaluates their impact on HRI in different time conditions.

Findings

High LOE is preferred when no time limit is imposed.

High and medium LOE improve task performance under time constraints.

Higher LOE reduces collisions, incorrect movements, and clarifications.

Abstract

In this work, we focused on constructing and evaluating levels of explanation(LOE) that address two basic aspect of HRI: 1. What information should be communicated to the user by the robot? 2. When should the robot communicate this information? For constructing the LOE, we defined two terms, verbosity and explanation patterns, each with two levels (verbosity -- high and low, explanation patterns -- dynamic and static). Based on these parameters, three different LOE (high, medium, and low) were constructed and evaluated in a user study with a telepresence robot. The user study was conducted for a simulated telerobotic healthcare task with two different conditions related to time sensitivity, as evaluated by two different user groups -- one that performed the task within a time limit and the other with no time limit. We found that the high LOE was preferred in terms of adequacy of explanation, number of collisions, number of incorrect movements, and number of clarifications when users performed the experiment in the without time limit condition. We also found that both high and medium LOE did not have significant differences in completion time, the fluency of HRI, and trust in the robot. When users performed the experiment in the with time limit condition, high and medium LOE had better task performances and were preferred to the low LOE in terms of completion time, fluency, adequacy of explanation, trust, number of collisions, number of incorrect movements and number of clarifications. Future directions for advancing LOE are discussed.