Reinforcement Learning Agent Trained as FES Controller in Modified Xia-Law Simulation Robustly Controls Grip Force.

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Neuroscience

Background:

• Functional electrical stimulation (FES) is effective for stroke and spinal cord injury rehabilitation.
• Clinical adoption of FES is limited by long setup times and complex parameter selection.
• Existing FES systems require significant expertise for optimal muscle activation.

Purpose of the Study:

• To develop an adaptive FES controller that simplifies parameter selection and reduces setup time.
• To improve the clinical usability and acceptance of FES therapy.
• To create an FES system robust to variations in stimulation parameters.

Main Methods:

• Trained a reinforcement learning agent using a novel FES-muscle activation model.
• Modulated stimulation amplitude to control muscle contraction.
• Tested the FES agent's grip force control accuracy across varying pulse widths (200 μs, 300 μs, 400 μs).
• Quantified performance using root-mean-square error (RMSE) against maximum evoked contraction (MEC).

Main Results:

• The FES agent achieved low RMSE across all tested pulse widths (3.50% MEC, 3.59% MEC, 3.45% MEC).
• Grip force control was robust and unaffected by the selected pulse width.
• Demonstrated consistent performance regardless of stimulation parameter choice.

Conclusions:

• The developed FES controller enhances accuracy and simplifies usage, addressing key barriers to clinical adoption.
• Reduced complexity and training requirements can increase the acceptance of FES therapy in rehabilitation settings.
• This technology has the potential to improve patient outcomes through more accessible FES rehabilitation.