Transformer-Based Approach for Predicting Transactive Energy in Neurorehabilitation.

Area of Science:

• Robotics
• Neurorehabilitation
• Biomedical Engineering

Background:

• Enhancing safety and personalization in physical human-robot interactions (pHRI) is crucial for robotic neurorehabilitation.
• Traditional control methods struggle with sensor noise and adaptability in unstructured environments.
• Energy transfer management is key for safe human-robot collaboration during rehabilitation.

Purpose of the Study:

• To introduce transactive energy as a coordinate-invariant metric for quantifying human-robot energy dynamics.
• To develop and validate a transformer-based deep learning model for predicting transactive potential energy.
• To enable personalized robot control in neurorehabilitation applications.

Main Methods:

• Developed a transformer-based deep learning model to predict transactive potential energy.
• Implemented the model on a compliant parallel ankle rehabilitation robot with three rotational degrees of freedom.
• Collected experimental data from five stroke patients using impedance and trajectory tracking controllers.

Main Results:

• The transformer model successfully learned from experimental data to predict transactive potential energy.
• Demonstrated the feasibility of using a deep learning approach for energy-based control in pHRI.
• Established a baseline for future research in energy-based control mechanisms.

Conclusions:

• Transactive energy offers a promising approach for enhancing safety and personalization in robotic neurorehabilitation.
• Transformer-based models can effectively estimate energy transfer dynamics in human-robot interactions.
• This work lays the foundation for advanced energy-aware control strategies in rehabilitation robotics.