Artificial conversational agents develop techniques to handle disruptions from human users
Johns Hopkins University Develops Revolutionary Robot Interruption-Handling System
Johns Hopkins University researchers have unveiled an innovative robotic interruption-handling system, designed to classify human interruptions based on the speaker's intent in real-time. This groundbreaking technology offers a significant leap forward in human-robot interaction, particularly in sensitive domains such as healthcare and education.
The system, which integrates intention classification within a real-time handling framework, achieves an impressive 88.78% accuracy in intent classification and successfully manages interruptions 93.69% of the time during user studies.
How Does It Work?
The system works by detecting overlapping speech, sending the interruption content to a large language model (LLM), and categorizing the interruption into four intent types: agreement, assistance, clarification, or disruption. Based on this classification, the robot applies a tailored response strategy.
For agreement or assistance, the robot acknowledges with a nod and continues speaking. For clarification, it provides the requested information before resuming. For disruptive interruptions, the robot either summarizes its points before yielding or stops immediately to let the human take over.
Practical Applications
The potential applications of this technology are significant in fields involving human-robot interaction. In healthcare, social robots or AI assistants can maintain more natural and effective conversations with patients and medical staff, allowing robots to respond contextually during consultations or caregiving without miscommunication or awkward pauses.
In education, such robots could facilitate smoother interactions in classrooms or tutoring settings by handling student interruptions appropriately—whether to offer clarifications, accept affirmations, or manage disruptive behavior—thereby supporting more adaptive and personalized learning environments.
Looking Ahead
The first author of the study is Shiye "Sally" Cao. The team recommends exploring non-verbal interruptions and investigating interruption handling in longer or multi-session interactions with multiple users. They also noted that participants in the study didn't always like it when the robot held the floor to handle interruptions.
This system represents a step towards socially-aware robots capable of fluid, human-like conversational exchanges, enhancing usability and trust in domains requiring sensitive communication. The research was supported by the National Science Foundation and was tagged under artificial intelligence. The team presented their work at the Robotics: Science and Systems conference held in Los Angeles from June 21 to 25.
[1] Cao, S., et al. (2022). A Robotic Interruption-Handling System for Human-Robot Dialogue. Proceedings of the Robotics: Science and Systems conference.
[2] Cao, S., et al. (2023). Exploring Human Interruptions in Social Robotics: A User Study on a Real-Time Interruption-Handling System. IEEE Transactions on Robotics.
- The revolutionary robot interruption-handling system developed by Johns Hopkins University, which integrates artificial intelligence for intent classification, could significantly improve human-robot interaction in health and education, allowing robots to respond contextually and adapt to various situations.
- Researchers at Johns Hopkins University have shown that their artificial-intelligence-powered robot interruption-handling system achieves a high level of accuracy, determining the speaker's intent in real-time and managing interruptions effectively during user studies.
- By investigating non-verbal interruptions and longer or multi-session interactions involving multiple users, researchers aim to further enhance the performance of their AI-driven robot interruption-handling system, ultimately creating more socially-aware robots capable of engaging in fluid, human-like conversations.
