Reinforcement Learning-Driven Path Generation for Ankle Rehabilitation Robot Using Musculoskeletal-Informed Energy...

Area of Science:

• Rehabilitation Robotics
• Biomechanics
• Machine Learning in Healthcare

Background:

• Rehabilitation robotics requires optimizing energy and interaction forces to avoid muscle fatigue and joint loading.
• Stroke patients often have asymmetrical muscle activation and impaired coordination, necessitating adaptive robotic systems.
• Inefficient trajectory planning can disrupt natural movement patterns during rehabilitation.

Purpose of the Study:

• To develop a reinforcement learning-based framework for trajectory optimization in a 3-DOF ankle rehabilitation robot.
• To integrate musculoskeletal modeling, transactive energy, and real-time physiological feedback for adaptive control.
• To enhance energy efficiency and control excessive torque for stroke patients' specific motor limitations.

Main Methods:

• Utilized a reinforcement learning framework for trajectory optimization.
• Integrated musculoskeletal modeling and transactive energy concepts.
• Incorporated real-time electromyography (EMG) signals and joint reaction forces for adaptive control.
• Validated the methodology with data from ten stroke patients.

Main Results:

• Generated adaptive rehabilitation trajectories tailored to individual patient needs.
• Demonstrated improved biomechanical efficiency through refined movement patterns.
• Showcased the system's ability to manage energy consumption and interaction forces effectively.

Conclusions:

• The developed framework shows potential for enhancing rehabilitation effectiveness in stroke patients.
• The system promotes more natural, efficient, and physiologically accurate movement trajectories.
• Adaptive trajectory optimization using reinforcement learning is a promising approach for personalized robotic rehabilitation.