Research on Robot Screwing Skill Method Based on Demonstration Learning.

Area of Science:

• Robotics
• Machine Learning
• Human-Robot Interaction

Background:

• Robots often struggle with generalization in tasks like screwing due to variations in scenarios and objects.
• Learning from human demonstrations is a promising approach to improve robot skill acquisition and adaptability.

Purpose of the Study:

• To develop and validate a robot screwing skill learning framework that enhances generalization ability across different scenarios and objects.
• To integrate human operational experience into a robot learning framework for improved task performance.

Main Methods:

• A task-based teaching, learning, and summarization framework was employed.
• Dynamic Movement Primitive (DMP) and Gaussian Mixture Model-Gaussian Mixture Regression (GMM-GMR) were utilized for skill learning and obstacle avoidance.
• The framework modeled hole-finding and screwing stages, incorporating potential functions for obstacle definition.

Main Results:

• The robot successfully avoided obstacles during tightening experiments.
• The developed framework demonstrated effectiveness in completing screwing tasks for various objects (bolts, caps, faucets).
• The robot tightening skill learning model showed adaptability to different tightening scenarios.

Conclusions:

• The proposed teaching-learning framework significantly improves robot generalization for screwing tasks.
• The integration of human experience and advanced machine learning techniques enhances robot adaptability and task completion in diverse environments.