Benchmark RadarREPAIR-Bench
REPAIR-Bench evaluates robot error perception and recovery in human-robot interaction. It includes 214 interaction trials from 41 participants with four induced failure types, synchronized facial action units, head pose, speech transcripts, and post-interaction reports. Three tasks cover failure detection across sessions, failure-type classification, and recovery prediction.
- Released
- 2026-06-29
- Readiness
- Paper only
- Primary field
- Robotics & Autonomous Systems
Why it matters
REPAIR-Bench addresses the lack of unified benchmarks for HRI failures, enabling standardized evaluation of failure detection, classification, and recovery prediction. This supports the development of adaptive and trustworthy robot systems and provides a comparison baseline for future research.
Motivation
Understanding how users perceive and respond to robot failures is essential for building robust and trustworthy robot systems.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.