Investigating muscle function during human movement relies on theoretical approaches because no reliable, humane method of estimating muscle forces in-vivo currently exists. In this study, a dynamic optimization solution for a full cycle of gait was computed using a detailed mathematical model of the human body. In contrast to previous dynamic models, the model in this study allowed unconstrained three-dimensional motion and was controlled by many muscles. To reduce the computational demands associated with solving such a problem, a parallel computational algorithm was formulated which, when profiled on IBM SP2 and Cray T3E computers, was found to reduce computational time by factors approaching 100 compared to serial computations.

The body was modeled as a 10-segment, 23-degree-of-freedom linkage which was free to make and break contact with the ground. The foot-ground interaction was simulated using a series of springs and dampers on the soles of the feet Each of the 54 musculotendinous units used to actuate the model was represented by a Hill-type contractile element in series with tendon. Active states of muscles were driven by a first-order differential equation relating activation to neural excitation. Physiological and geometric parameters of muscles were based on the work of Delp (1990). Model inertial properties were set by averaging the anthropometries of five male subjects. Model muscle strengths were scaled based on torque-angle data from those same subjects.

Dynamic optimization problems for maximum-height jumping and for normal gait were solved. The jumping solution was used to validate and refine the musculoskeletal model. The performance criterion for the gait solution was chosen to be the minimization of metabolic energy expended per unit distance traveled. The gait solution was validated through comparison with subject data. Muscle induced accelerations computed from the solution revealed dominant muscle actions during gait.

Comparison of the dynamic solution for gait to two simpler static optimization solutions suggested the great computational expense of the former is justified when:​ 1) accurate experimental data are not available, 2) activation dynamics is suspected to be significant, 3) an appropriate static performance criterion cannot be posed, or 4) predictive simulation of movement is desired.

This work represents a significant contribution to the study of human movement The musculoskeletal model developed is a reusable resource suitable for simulating many activities beyond jumping and gait The dynamic solution for gait provides guidelines for deciding when the computational cost of dynamic optimization is justified. The parallel computation algorithm renders complex dynamic problems solvable and enables predictive dynamic simulation. The tabulation of muscle actions provides clinicians with a valuable resource for enhancing treatment of gait pathologies.