“I'm Done”: Describing Human Reactions to Successive Robot Failure

Abstract

Robots are imperfect and will often fail multiple times during interactions. Despite this, there is a knowledge gap in understanding how humans respond to successive robot failures. In a user study with 26 participants, we explored human responses to successive robot conversational errors. We found that users typically resort to reformulating their prompts, modifying the verbal tone and cadence when encountering repeated failures. A range of emotional displays emerges, including confusion, frustration, and occasional amusement, which evolve throughout the interaction. We further conducted a statistical analysis of participants' behavioral features, revealing insights that could inform automated approaches to detecting and responding to robot errors in successive failure scenarios.