Strategic Surprise: Micro-Failures as a Mechanism to Calibrate Shared Mental Models

Wróbel et al. (2023, RO-MAN) categorize how children respond to six kinds of surprising robot behaviors (personality deviations, movement malfunctions, inconsistency, mispronunciation, delays, freezing), finding clear age-dependent patterns. Rehm and Krogsager (2013, RO-MAN) show that unexpectedness can trigger negative affect and impoliteness in adults. Rather than treating unexpected behaviors purely as errors to be eliminated, this project treats them as carefully designed “diagnostic perturbations”—short, low-stakes departures from expectations that let the robot actively learn a user’s tolerance, repair preferences, and interpretations. The robot triggers a micro-failure, then observes the human’s verbal/nonverbal responses and recovery dynamics, using Mizuchi et al.’s (2023, HAI) human-centered evaluation factors (e.g., instruction conciseness, adaptive guidance) to infer how to adjust its future interaction strategy. To avoid over-reliance on LLM “theory of mind” (which can look good but fail under perturbations; Verma et al., 2024, HRI), the robot maintains an explicit, symbolic user-expectation model that is updated via causal inference from micro-failure outcomes. What’s new is using intentional, bounded surprise as an online model-learning tool, grounded in an empirically derived taxonomy of unexpected behaviors and age differences (Wróbel et al., 2023), rather than as noise to be filtered. This could yield robust personalization with minimal data, reduce brittle overfitting to normative scripts, and improve safety by identifying low tolerance to failure early. Impact: a principled way for robots to become better partners faster—especially with children and non-experts—by “asking questions” through designed, recoverable deviations.

References:

1. Age-Appropriate Robot Design: In-The-Wild Child-Robot Interaction Studies of Perseverance Styles and Robot’s Unexpected Behavior. Alicja Wróbel, Karolina Źróbek, Marie-Monique Schaper, Paulina Zguda, B. Indurkhya (2023). IEEE International Symposium on Robot and Human Interactive Communication.
2. Negative affect in human robot interaction — Impoliteness in unexpected encounters with robots. M. Rehm, Anders Krogsager (2013). 2013 IEEE RO-MAN.
3. Designing Evaluation Metrics for Quality of Human-Robot Interaction in Guiding Human Behavior. Y. Mizuchi, Yusuke Tanno, T. Inamura (2023). International Conference on Human-Agent Interaction.
4. Theory of Mind Abilities of Large Language Models in Human-Robot Interaction: An Illusion?. Mudit Verma, Siddhant Bhambri, Subbarao Kambhampati (2024). IEEE/ACM International Conference on Human-Robot Interaction.