REX: Designing User-centered Repair and Explanations to Address Robot Failures

TL;DR

The paper examines how autonomous repair actions and explanatory communication (REX) can improve user experience when robots face conflicts in real-world, multi-user settings. Through two studies—an online vignette-based study with 162 participants and an in-person study with 24 participants using a physical robot—the work shows that automated repairs and detailed explanations boost trust, satisfaction, and perceived usefulness, while identifying safety, privacy, and complexity as key risk factors requiring adaptive strategies. It develops design insights and guides for implementing REX, including when to involve users, how to structure explanations, and how to tailor repairs to risk type and severity. The findings offer a roadmap for building user-centered, robust robot systems capable of handling diverse, dynamic environments with multiple users.

Abstract

Robots in real-world environments continuously engage with multiple users and encounter changes that lead to unexpected conflicts in fulfilling user requests. Recent technical advancements (e.g., large-language models (LLMs), program synthesis) offer various methods for automatically generating repair plans that address such conflicts. In this work, we understand how automated repair and explanations can be designed to improve user experience with robot failures through two user studies. In our first, online study ($n=162$), users expressed increased trust, satisfaction, and utility with the robot performing automated repair and explanations. However, we also identified risk factors -- safety, privacy, and complexity -- that require adaptive repair strategies. The second, in-person study ($n=24$) elucidated distinct repair and explanation strategies depending on the level of risk severity and type. Using a design-based approach, we explore automated repair with explanations as a solution for robots to handle conflicts and failures, complemented by adaptive strategies for risk factors. Finally, we discuss the implications of incorporating such strategies into robot designs to achieve seamless operation among changing user needs and environments.

Figures (6)

• Figure 1: Study Materials for Online User Study ---The table shows the six vignettes and the repair and explanation conditions used in the online user study. The vignettes were derived from an HRI failure taxonomy.
• Figure 2: Quantitative Data from Online User Study ---Box plots with overlaid data points on participant trust and satisfaction scores across different conditions. Horizontal lines indicate significant pairwise comparisons with Tukey's HSD ($p < 0.05^{\ast}$, $p < 0.01^{\ast\ast}$, $p < 0.001^{\ast\ast\ast}$).
• Figure 3: Quantitative Data from Online User Study ---Box plots with overlaid data points on participant trust and satisfaction scores across different scenarios. Horizontal lines indicate significant pairwise comparisons with Tukey's HSD ($p < 0.05^{\ast}$, $p < 0.01^{\ast\ast}$, $p < 0.001^{\ast\ast\ast}$).
• Figure 4: Study Procedure for In-person User Study --- The figure depicts a participant engaging in the in-person user study. Left: While preoccupied with a task, the participant delegates a different task to the robot. Middle: The robot attempts to fulfill the user's request but encounters an unexpected conflict. Its repair actions are displayed in the vignette. Right: After completing their original task, the participant observes the robot's actions. Finding these actions unexpected, the participant verbally interacts with the robot to obtain explanations for its actions and rationale.
• Figure 5: Study Materials for In-person User Study ---The table presents the three vignettes along with the repair and explanation conditions used in the in-person user study. These vignettes were designed based on the risk factors (i.e., safety, complexity, and privacy) identified in the initial online user study.