Transformer-Based Modular Motion Generation Through Shoulder-Arm Decoupling for Upper Limb Rehabilitation.

Area of Science:

• Robotics and Biomechanics
• Machine Learning in Healthcare
• Rehabilitation Engineering

Background:

• Modeling upper limb biomechanics, especially the shoulder girdle and arm coupling, presents significant challenges.
• Existing methods often struggle with subject-specific variability and generalization in rehabilitation scenarios.
• Accurate human motion prediction is crucial for effective personalized rehabilitation strategies.

Purpose of the Study:

• To propose a modular motion-planning framework that explicitly decouples the shoulder girdle from the arm for improved upper limb motion modeling.
• To enhance the adaptability and interpretability of motion prediction models for rehabilitation.
• To develop a method for translating predicted human motion into control signals for rehabilitation exoskeletons.

Main Methods:

• A Transformer-based deep learning architecture was utilized for human motion modeling, capturing complex joint angle and scapular dynamics.
• The framework separates shoulder girdle prediction from arm trajectory integration, allowing independent optimization.
• A machine learning approach mapped predicted human motion to a 6-DOF rehabilitation exoskeleton's configuration space.

Main Results:

• The Transformer model demonstrated superior accuracy and temporal alignment compared to experimental data, reinforcement learning, and long short-term memory models.
• Quantitative analyses using F1-Score, ANOVA, T-test, RMSE, and DTW confirmed the model's performance.
• The modular approach proved resilient to subject-specific variability, enhancing scalability and personalization.

Conclusions:

• The proposed framework effectively models and predicts upper limb motion by decoupling the shoulder girdle and arm.
• Transformer-based deep learning offers a powerful tool for capturing nonlinear biomechanical dependencies in human movement.
• This approach provides a clinically relevant and adaptable solution for personalized upper limb rehabilitation using exoskeletons.