Transparency Mechanisms to Improve Robot Failure Identification

Goal

Test if robot transparency mechanisms improve people's mental models of the robot

Research Questions

  1. Do transparency mechanisms help users identify the cause of the robot's failure?
  2. What is the effect of transparency mechanisms on the user's mental model of the robot?

Tasks of Focus

12 different failure scenarios in a simulated tabletop robot setup

Methods

Usability testing

Analysis Methods

Statistical Methods, Inductive Thematic Analysis

Population

64 participants recruited on Amazon Mechanical Turk

Select study results

  • Participants were more accurate at identifying the robot's failure when they used question-answering, or both question-answering and visual mechanisms compared to when they had no access to transparency mechanisms.
  • Even when participants had access to transparency mechanisms, even though their mental models of the robot were better, they were still not accurate.

Paper

Preeti Ramaraj, Saurav Sahay, Shachi H. Kumar, Walter S. Lasecki, and John E. Laird. "Towards using transparency mechanisms to build better mental models." In Advances in cognitive systems: 7th goal reasoning workshop, vol. 7, pp. 1-6. 2019.