Gesture Interaction Failures in Public-Space Kiosks: A Failure Analysis
A new paper on arXiv explores the gap between gesture recognition accuracy and how well these gestures work in public interactive systems. It examines eight engineering logs from a kiosk project in a picturesque location, focusing on four specific gestures: two-handed bowing, one-handed fist shaking, two-handed catching, and hovering over knowledge-graph nodes. Researchers found 20 failure cases, which they grouped into six overlapping categories, including issues like model-output degeneration and temporal mismatches. The study highlights that interactive kiosks and service terminals encounter different challenges than what is typically seen in labs, stressing the need to address the 'recognition-to-interaction gap' in future designs.
Key facts
- arXiv:2607.21601v1 analyzes gesture interaction failures in public-space kiosks.
- 8 engineering repair records from a scenic-area interactive kiosk project were studied.
- 4 gesture tasks were covered: two-hand bowing, single-hand fist shaking, two-hand catching control, and knowledge-graph node hovering.
- 20 failure instances were extracted from the repair traces.
- Six non-exclusive working failure classes were identified.
- The failure classes include model-output degeneration, temporal mismatch, geometric-scale instability, coordinate-rendering mismatch, runtime lifecycle failure, and feedback synchronization and recovery failure.
- The paper identifies a 'recognition-to-interaction gap' between frame-level recognition and event-level confirmation.
- Deployed systems in scenic kiosks, exhibition halls, and service terminals expose different failure boundaries than lab evaluations.
Entities
Institutions
- arXiv