Calling for Backup: How Children Navigate Successive Robot Communication Failures
1Cornell University · 2 Lasige, Faculdade de Ciências · 3 ITI/LarSys, Universidade de Lisboa· 4Cornell Tech
*Equal contribution
TBD
Reactions to successive robot error. After providing instructions to the child, the robot fails to understand their prompt three times.
Design Insights for Child-Robot Interaction
| Observed Behavior | Design Implication |
|---|---|
| Children's perception remained stable despite successive errors | Design for Error Tolerance: Focus on managing errors gracefully rather than aggressively eliminating them. Allow minor errors without elaborate apologies. |
| Children frequently call for adult help after repeated failures | Facilitate External Help-Seeking: Recognize requests for human assistance as valid and implement graceful handoff mechanisms. |
| Children use sophisticated verbal strategies (repetition, specificity, politeness) | Implement Multi-Layered Conversational Repair: Design systems sensitive to prompt reformulation and increase error confidence based on politeness markers. |
| Children exhibit dynamic engagement patterns | Employ Non-Intrusive Error Recovery: Recognize disengagement as active problem-solving. Avoid intrusive re-engagement; monitor passively. |
| Response latencies increase with successive errors | Utilize Timing as Failure Metric: Integrate response latency analysis as real-time measure of cognitive load and distress. |
Abstract
How do children respond to repeated robot errors? While prior research has examined adult reactions to successive robot errors, children's responses remain largely unexplored. In this study, we explore children's reactions to robot social errors and performance errors. For the latter, this study reproduces the successive robot failure paradigm of Liu et al. with child participants (N=59, ages 8-10) to examine how young users respond to repeated robot conversational errors.
We found both similarities and differences compared to adult responses. Like adults, children adjusted their prompts, modified their verbal tone, and exhibited increasingly emotional non-verbal responses throughout successive errors. However, children demonstrated more disengagement behaviors, including temporarily ignoring the robot or actively seeking an adult. Errors did not affect participants' perception of the robot, suggesting more flexible conversational expectations in children.
Research Question
How do children perceive and react to repeated robot error?
To investigate this question, we examine children's responses to two types of robot failures:
Social Error (Interruption)
Robot inappropriately interrupts the child during conversation, violating social norms.
Performance Error (Successive Failure)
Robot fails to understand the child's request to "call the researcher" three successive times.
Study Overview
Experimental setup (A) and study protocol. After a short introduction to robot Simon, children watch a series of videos on the laptop. Following this, they fill out a small survey evaluating different dimensions of robot perception. Once they finish, Simon asks them some questions: depending on the condition, it will interrupt their answers (social error) or not. After this, Simon fails to understand the participants’ request to call the researcher (performance error). After 3 successive errors, the researcher is called back into the room. Before ending the session, the child repeats the robot perception survey.
Participants: 59 children, ages 8-10 (32 boys, 27 girls), recruited through local school network in Portugal.
Design: Between-subjects design with two conditions:
- Interruption (n=30): Robot interrupts child before they finish answering
- Control (n=29): Robot does not interrupt
- All participants experienced 3 successive performance errors
Protocol: ~7 minute session including: (1) Introduction to robot, (2) Video visualization task, (3) Pre-interaction survey, (4) Social error manipulation, (5) Performance error (3 failures), (6) Post-interaction survey, (7) Debriefing.
Key Findings
Social Error: Interruption
- 84% of children (21/25) shifted their answer to the new question, not acknowledging the interruption
- Only 2 children continued responding to the previous question
- Children appeared largely oblivious to interruption as a social violation
Performance Error: Similarities with Adults
Like adults (Liu et al., 2025), children demonstrated:
- Verbal adaptation: Modified prompts and adjusted vocal tone/cadence
- Emotional progression: Evolution from confusion to frustration
- Response timing: Increased latencies with successive failures
- Interaction abandonment: Some discontinued interaction by third error
Performance Error: Differences from Adults
- Greater disengagement: More behaviors like ignoring the robot
- "Calling for Backup": Unlike adults, children sought help from researcher
- Politeness strategies: Shifted to "please" when commands failed
- Earlier disengagement: Started at Error I vs. adults at Error III
- Mixed patterns: Dynamic disengagement/re-engagement cycles
Behavioral Response Categories
Observed behaviors across 402 coded annotations:
- Verbal reprompting: Repeating, more specific/longer prompts, simpler prompts, term swapping
- Vocal modifications: Slower speech, demanding/interrogative/assertive tone
- Physical behaviors: Moving closer to robot, standing up, leaving room
- Emotional displays: Confusion, frustration, amusement/humor
- Disengagement: Looking for researcher, no prompt, quitting interaction
Most common engagement trajectories (Error I → II → III):
- Engage → Engage → Engage (n=15)
- Engage → Disengage → Engage (n=5)
- Engage → Disengage → Disengage (n=4)
Robot Perception
No significant changes in children's perception of the robot across any dimension (willingness to interact, competence, trust, social acceptance, likeability). No differences between Interruption and Control conditions (Mann-Whitney U tests, Bonferroni-corrected α = 0.01). This suggests children have more flexible conversational expectations with robots compared to adults.
Data Visualizations
Distribution of children's behavioral responses across successive robot errors. Shows the relative frequency of different reaction types.
Frequency of emotional reactions across successive robot errors. Confusion peaks after Error II, while frustration progressively increases and becomes dominant by Error III. Amusement/humor remains relatively consistent across all errors.
Response latencies across successive robot errors. Response times increase and become more variable with successive errors, with Error III showing the longest delays and greatest variability.
Children's engagement patterns across successive robot errors. Blue bars represent continued robot engagement, green bars show disengagement from robot interaction, pink and yellow show complex engagement patterns. Notable shift from predominantly engaging behavior in Error I to mixed patterns in Errors II and III.
Frequency of disengagement behaviors across successive robot errors. Shows progressive increase in disengagement strategies, with "no prompt" and "looks for researcher" being most prevalent by Error III. Other behaviors include standing up, leaving room, and complete interaction abandonment.
Main Takeaways
- Children exhibit unique error recovery patterns: Unlike adults who persist with self-directed problem-solving, children actively seek external help from adults when robots fail repeatedly.
- Flexible expectations protect perception: Robot errors did not negatively impact children's perception across any measured dimension, suggesting developmental advantages in maintaining positive robot relationships despite failures.
- Dynamic engagement is adaptive: Children's cycling between engagement and disengagement represents sophisticated meta-cognition about tool limitations and alternative problem-solving strategies.
- Social errors go unnoticed: Most children ignored robot interruptions, suggesting either developmental differences in social norm sensitivity or more flexible conversational expectations with non-human agents.
Citation
@inproceedings{parreira2026calling,
title={Calling for Backup: How Children Navigate Successive Robot Communication Failures},
author={Parreira, Maria Teresa and Neto, Isabel and Rocha, Filipa and Ju, Wendy},
booktitle={TBD},
year={2026},
}