Motion-generation system for violin-playing robot using reinforcement learning differences in bowing parameters due...

Area of Science:

• Robotics
• Artificial Intelligence
• Music Technology

Background:

• Human-robot communication is a growing research area.
• Music serves as a vital channel for conveying emotional information between humans and robots.
• Automating musical performance, like robotic violin playing, enhances understanding of human-robot musical interaction.

Purpose of the Study:

• To develop a system for a violin-playing robot to automatically determine bowing parameters (speed, direction) from musical scores.
• To enable expressive sound production through learned performance nuances.
• To utilize reinforcement learning for optimizing robotic musical expression.

Main Methods:

• Employed Q-learning and ε-greedy methods for reinforcement learning.
• Utilized a neural network to approximate the value function.
• Integrated sound pressure values into musical scores to guide bowing parameter determination.

Main Results:

• The system successfully learned to determine bowing parameters for expressive violin performance.
• Adjusting learning conditions and target sound pressure values improved performance.
• Negative rewards decreased, and positive rewards increased, indicating successful learning.

Conclusions:

• The developed system enables a violin-playing robot to automatically interpret musical scores and produce expressive sounds.
• Reinforcement learning is a viable method for achieving nuanced robotic musical performance.
• This research advances human-robot communication through expressive musical interaction.