Establishing Personalized and Sparse Spinal Reflex Circuitry from Locomotion Data.

Area of Science:

• Biomechanics and Motor Control
• Computational Neuroscience
• Robotics and Control Systems

Background:

• Spinal reflex circuitry controllers show potential for simulating human locomotion.
• Previous research often predefined reflex structures, limiting generalizability and understanding of human locomotion.
• Identifying dominant reflex loops is crucial for advancing locomotion control.

Purpose of the Study:

• To propose an identification framework for personalized, sparse-structure spinal reflex circuitry directly from experimental data.
• To overcome limitations of predefined reflex structures in previous studies.
• To gain a better understanding of dominant reflex loops in human locomotion.

Main Methods:

• Developed a framework for identifying personalized sparse-structure reflex circuitry.
• Employed muscle model personalization via optimization and extensive model regularization for neural reflex networks.
• Utilized trajectory optimization (direct collocation as a non-linear program) for both personalization and regularization steps.
• Tested the framework using multi-speed walking and running data from five subjects.

Main Results:

• Successfully identified personalized sparse reflex circuitry that accurately reproduces torque and muscle activation across multiple walking speeds.
• The identified reflex network was speed-independent but phase-dependent.
• Found similar sparse structures across subjects, with distinct structures for walking versus running gaits.

Conclusions:

• The proposed framework automatically identifies personalized sparse reflex circuitry, advancing locomotion control.
• The identified circuitry demonstrates exceptional reproduction of biomechanical data, suggesting a robust control mechanism.
• While plausible, the exact biological relevance of the found circuitry requires further investigation, particularly concerning unperturbed locomotion data.