Reinforcement Learning Identifies Age-Related Balance Strategy Shifts.

Area of Science:

• Biomechanics
• Gerontology
• Computational Neuroscience

Background:

• Falls are a major cause of injury and death in the elderly.
• Sarcopenia, the age-related loss of muscle mass, contributes to muscle weakness and impaired balance.
• Direct human experimentation on balance and sarcopenia faces ethical and cost limitations.

Purpose of the Study:

• To investigate the effects of sarcopenia on standing balance using neuromusculoskeletal modeling.
• To develop and validate a computational model that simulates balance control in elderly individuals with varying degrees of sarcopenia.

Main Methods:

• A novel full-body neuromusculoskeletal model with 290 muscle actuators and 23 degrees of freedom was created.
• Reinforcement learning, curriculum learning, and muscle synergy representations were employed to train a balance control agent.
• The model simulated standing balance on a backward-sliding plate, comparing healthy and sarcopenic agent behaviors to human experiments.

Main Results:

• The computational model successfully reproduced human ankle and hip balancing strategies for both healthy and sarcopenic agents without using experimental data.
• Simulated agents demonstrated adaptive balancing strategies, adjusting to platform acceleration based on the severity of sarcopenia.
• The study highlights the potential of computational modeling to understand age-related balance deficits.

Conclusions:

• Neuromusculoskeletal modeling provides a viable alternative to human experimentation for studying sarcopenia's effect on balance.
• The developed model accurately simulates balance control mechanisms and adaptive strategies in the face of muscle weakness.
• This research offers insights into fall prevention and rehabilitation strategies for the elderly population.