ExGenNet: Learning to Generate Robotic Facial Expression Using Facial Expression Recognition

Area of Science:

• Robotics
• Artificial Intelligence
• Human-Robot Interaction

Background:

• Generating appropriate robot facial expressions is crucial for perceived sociability.
• Current methods often use fixed, preprogrammed joint configurations, limiting scalability and adaptability.
• Automating expression generation offers advantages for diverse robot types and expressions.

Purpose of the Study:

• Introduce ExGenNet, a novel deep generative approach for facial expression generation on humanoid robots.
• Enable automated, scalable, and transferable facial expression generation.
• Optimize robot joint configurations for various expressions.

Main Methods:

• ExGenNet utilizes a generator network to reconstruct facial images from robot joint configurations and a classifier network for expression recognition.
• Joint configurations are optimized by backpropagating the classification loss through both networks.
• Feature extraction in both networks improves transferability between human training data and different robot appearances.

Main Results:

• ExGenNet successfully generated sets of joint configurations for predefined facial expressions on two distinct robots (Alfie and Elenoide).
• The system demonstrated the ability to produce multiple configurations for each facial expression.
• Generated expressions were transferable between different robot platforms.

Conclusions:

• ExGenNet provides an effective automated method for generating realistic facial expressions on humanoid robots.
• The generated expressions were accurately recognized by human subjects, validating the system's effectiveness.
• The approach enhances human-robot interaction by improving the perceived sociability of robots.