A Novel Optimization Approach to Generate Physiological Human Walking Patterns

Dynamic simulation of the musculoskeletal system is increasingly being used to study human normal and pathological walking. The common approach to predict walking patterns is based on the assumption that the central nervous system minimizes an intrinsic performance criterion. For instance, during walking, the energy expenditure per unit of distance traveled was shown to play a key role. The resulting optimal control problem is almost exclusively solved by the so-called dynamic optimization. Dynamic optimization relies on the parameterization of neural excitations using nodal values serving as optimization variables. The reconstructed neural excitations are then used to numerically integrate the differential equations describing the dynamics of the musculoskeletal system. This approach has been successfully applied to predict salient normal walking patterns, including muscle coordination and energy expenditure. In spite of the growing use of dynamic optimization, the extremely high computational effort arising from the several numerical integrations of the large-scale state equations required prevents it from being more widely applied, e.g., for bioassistive devices. Approaches based on inverse dynamics have the potential to reduce the high computation effort by avoiding the necessity of numerically integrating the state equations, but have been poorly explored in biomechanics. The development of an inverse dynamics-based approach to generate near-optimal human walking patterns that deals with the overdeterminacy of muscle actuation in conjunction with Hill-type muscle models widely used in biomechanics is proposed in this paper. The approach is based on the parameterization of the motion and muscle forces. The neural excitations are obtained by inverting the muscle contraction and activation dynamics. The compatibility between motion and muscle forces is guaranteed by checking the fulfillment of the equations of motion of the skeletal system at control points. The approach is implemented and human normal and pathological gaits are generated and applied to the design of transtibial prostheses.