Bilevel Optimization for Cost Function Determination in Dynamic Simulation of Human Gait.

Area of Science:

• Biomechanics and Human Movement Science
• Computational Modeling and Simulation
• Robotics and Control Systems

Background:

• Predictive musculoskeletal simulation is crucial for understanding human gait dynamics.
• Current methods often assume the cost function a priori, limiting accuracy.
• Inverse optimal control problems require determining the underlying cost function.

Purpose of the Study:

• To develop and validate a bilevel optimization approach for solving inverse optimal control problems in musculoskeletal modeling.
• To determine the unknown cost function for predictive musculoskeletal simulations of human gait.
• To identify optimal performance criteria weights for normal human walking.

Main Methods:

• A nested evolutionary approach was proposed to solve the bilevel optimization problem.
• The lower-level optimization utilized a direct collocation method.
• The upper-level optimization employed a genetic algorithm.

Main Results:

• The proposed nested evolutionary approach effectively solved bilevel optimization problems.
• Demonstrated ability to find weights for common performance criteria in human walking cost functions.
• The method proved effective for determining underlying cost functions in predictive simulations.

Conclusions:

• The developed approach offers practical utility for designing assistive mobility devices.
• Provides valuable insights into the neural control mechanisms underlying human walking.
• Enables more accurate predictive musculoskeletal simulations by identifying optimal cost functions.