Automatic Aesthetics Evaluation of Robotic Dance Poses Based on Hierarchical Processing Network.

Area of Science:

• Robotics
• Artificial Intelligence
• Computer Vision
• Neuroaesthetics

Background:

• Human dancers use visual feedback for aesthetic evaluation and performance improvement.
• Robots require similar visual perception mechanisms for imitating human dance behavior.
• Existing methods lack effective automatic aesthetic evaluation for robotic dance poses.

Purpose of the Study:

• To develop an artificial intelligence system enabling robots to visually perceive and evaluate their own dance poses.
• To construct a novel dataset of robotic dance poses using NAO robots.
• To propose a hierarchical processing network for automatic aesthetic evaluation of robotic dance poses.

Main Methods:

• A hierarchical processing network inspired by neuroaesthetics.
• Utilizing three parallel Convolutional Neural Networks (CNNs) for primary visual feature extraction.
• Employing a synthesis CNN with multi-modal feature fusion for high-level processing and aesthetic decision-making.
• Creating a new dataset of dance poses from real NAO robots.

Main Results:

• The proposed hierarchical processing network achieved a high correct ratio of 82.3% in aesthetic evaluation.
• The method demonstrated superior performance compared to existing approaches.
• The novel dataset provides a foundation for robotic dance pose research.

Conclusions:

• The developed approach enables robots to automatically evaluate the aesthetic quality of their dance poses.
• This system mimics human visual feedback mechanisms for performance enhancement in robots.
• The findings contribute to advancing artificial intelligence in creative domains like dance choreography.