Walking representation and simulation based on multi-source image fusion and multi-agent reinforcement learning for...

Area of Science:

• Biomedical Engineering
• Rehabilitation Science
• Artificial Intelligence

Background:

• Accurate identification of current walking states and prediction of future states are critical for effective walking rehabilitation strategies.
• Existing methods for motion capture and gait simulation often fall short in precision and predictive capability.

Purpose of the Study:

• To develop and validate a unified framework for walking rehabilitation.
• To integrate a novel 3D walking motion capture method with multi-source image fusion.
• To implement a walking rehabilitation simulation approach using multi-agent reinforcement learning.

Main Methods:

• A novel 3D walking motion capture technique utilizing multi-source image fusion.
• A multi-agent reinforcement learning approach for gait rehabilitation simulation.
• Comparative analysis against visual skeleton tracking and IMU-based methods.

Main Results:

• The proposed 3D motion capture method achieved high accuracy (Pearson's r=0.93, ICC=0.91) compared to existing techniques.
• The multi-agent reinforcement learning simulation demonstrated convergence to target states and mimicked clinical gait rehabilitation patterns.
• The framework showed strong correlation and concurrence with IMU-based gait parameter assessment.

Conclusions:

• The developed framework provides precise 3D walking motion capture and effective gait rehabilitation simulation.
• This integrated approach has the potential to significantly improve clinical practice and patient outcomes in walking rehabilitation.