Impact of Different Failures on a Robot's Perceived Reliability

Understanding how users perceive and respond to robot failures is essential for building robust and trustworthy robot systems. Prior work, however, (i) often treats failures as independent events, (ii) emphasizes binary failure detection, (iii) with rule-based recovery modeling. We present REPAIR-Bench, built on 214 interaction trials from 41 participants, the benchmark spans four induced failure types and provides synchronized facial action units, head pose, speech transcripts, and post-interaction affect and recovery reports. The benchmark spans three novel evaluation tasks that jointly capture the lifecycle of failure in human-robot interaction (HRI): (i) failure detection over inter-dependent interaction sessions, modeling longitudinal user adaptation across repeated failures; (ii) visual failure-type classification beyond binary success/failure formulations; and (iii) user-centered recovery prediction, inferring users' preferred recovery strategies from interaction context rather than relying on manually designed or rule-based strategies. In baseline experiments, hierarchical recurrent modeling improved failure detection over a single-session model (strict F1: 0.80 vs. 0.68), achieved a failure localization mean signed error of -0.51 s, median absolute error of 2.97 s and, for recovery prediction, a QLoRA-tuned Mistral-7B reached Hit@5=0.76 and F1@5=0.32. REPAIR-Bench provides both the HRI and Medical HRI communities with a standardized framework for (1) evaluating robot failures and (2) building transparent, adaptive, and trustworthy recovery systems.

While robots deployed in real-world environments inevitably experience interaction failures, understanding how users respond through verbal and non-verbal behaviors remains under-explored in human-robot interaction (HRI). This gap is particularly significant in healthcare-inspired settings, where interaction failures can directly affect task performance and user trust. We present the Robot Failures in Medical HRI (RFM-HRI) Dataset, a multimodal dataset capturing dyadic interactions between humans and robots embodied in crash carts, where communication failures are systematically induced during item retrieval tasks. Through Wizard-of-Oz studies with 41 participants across laboratory and hospital settings, we recorded responses to four failure types (speech, timing, comprehension, and search) derived from three years of crash-cart robot interaction data. The dataset contains 214 interaction samples including facial action units, head pose, speech transcripts, and post-interaction self-reports. Our analysis shows that failures significantly degrade affective valence and reduce perceived control compared to successful interactions. Failures are strongly associated with confusion, annoyance, and frustration, while successful interactions are characterized by surprise, relief, and confidence in task completion. Emotional responses also evolve across repeated failures, with confusion decreasing and frustration increasing over time. This work contributes (1) a publicly available multimodal dataset (RFM-HRI), (2) analysis of user responses to different failure types and preferred recovery strategies, and (3) a crash-cart retrieval scenario enabling systematic comparison of recovery strategies with implications for safety-critical failure recovery. Our findings provide a foundation for failure detection and recovery methods in embodied HRI.

A robot that fails at a task faces the first decision in corrective dialogue: act on its own diagnosis, consult another onboard sensor, or interrupt a person. Choosing well requires knowing how much the robot's sensors reveal about the cause and how reliable the robot's own diagnosis is. We build a simulated benchmark in which every failure's true cause is known, because we injected it, and measure what each sensor reveals, with explicit checks against data leakage. Some failures are diagnosable from camera images; others only from the robot's force data (0.99 from force data, no image method above 0.55). We then test six open vision-language models. Their behavior tracks the surface of the prompt, not the evidence: moving the refusal option from last to first in the answer list collapses refusal rates from 78-100% to 0-6% in three of the six swept model-and-family pairs. Accuracy from frames stays at or below a majority-class baseline under every prompt variant, with or without worked examples, and stated confidence carries no information about correctness. Handing the same models the force data as ten lines of text produces the first above-baseline diagnoses, in four of the six models: much of the failure reflects missing sensor data, not missing ability. We pose the choice as a three-action decision problem, act, consult your own sensors, or ask a human, whose optimal policy follows from measured accuracy. The models do not follow it, and their ask rates ignore a fourfold change in question cost. One question to a human still lifts them from that baseline to roughly the answerer's own reliability (0.70-0.81 when they ask). The decision to ask should be tied to measured accuracy and stated costs, not to the model's confidence.