Toward an Attentive Robotic Architecture: Learning-Based Mutual Gaze Estimation in Human-Robot Interaction.

Area of Science:

• Robotics
• Human-Robot Interaction
• Computer Vision

Background:

• Social robotics is a rapidly growing field with increasing human-robot collaboration.
• Endowing humanoid robots with human-like social behaviors, particularly gaze, remains a challenge.
• Mutual gaze is a critical social cue for natural face-to-face interaction.

Purpose of the Study:

• To implement a mechanism for estimating gaze direction in humanoid robots.
• To automatically detect eye contact events during online human-robot interactions.
• To develop a foundational component for attentive robot architectures.

Main Methods:

• A learning-based framework was developed for gaze estimation.
• The system automatically detects eye contact events.
• Performance was validated through both simulated (in silico) and experimental testing.

Main Results:

• The proposed framework achieved high performance in detecting eye contact.
• The solution demonstrated effectiveness in both simulated and real-world scenarios.
• Successful implementation of gaze estimation for human-robot interaction.

Conclusions:

• The developed system is a significant step towards robots perceiving social cues.
• This research contributes to creating robots that can be perceived as social partners.
• The framework lays the groundwork for more sophisticated attentive robot architectures.