Are You Still With Me? Continuous Engagement Assessment From a Robot's Point of View.

Area of Science:

• Robotics
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Measuring user engagement in human-robot interaction (HRI) is crucial for improving interaction quality and guiding robot behavior.
• Existing methods struggle to capture the multifaceted nature of engagement in a generalized computational model.
• Robots need a reliable way to assess user engagement for real-time learning and adaptation.

Purpose of the Study:

• To develop a novel computational model for robots to continuously measure user engagement during HRI.
• To create a single scalar metric for engagement using standard video streams from the robot's perspective.
• To enable robots to understand and adapt to user engagement levels for optimized interactions.

Main Methods:

• A novel regression model utilizing Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks was developed.
• The model was trained on a long-term dataset from an autonomous tour guide robot in a public museum.
• Engagement was continuously annotated by three independent coders to provide ground truth data.

Main Results:

• The proposed model accurately predicts user engagement in the original application domain (museum tour guide).
• The model demonstrated successful transferability to a different dataset with varying tasks, environments, and participants.
• The developed model provides a robust and generalizable method for quantifying engagement in HRI.

Conclusions:

• The developed CNN-LSTM regression model offers a viable solution for robots to computationally assess user engagement.
• The model's ability to generalize across different HRI scenarios highlights its potential for broad application.
• The availability of the model and software promotes further research and development in HRI engagement measurement.