To provide a parameter optimization framework within the OpenSim API (and eventually within the GUI) for performing optimal control simulations of human movement.
Optimal control models of human movement are attractive for their ability to elegantly simulate complex movements by minimizing comparatively simple cost functions. Concepts from optimal control theory have been used in human movement science for decades, and models of this sort have arguably been more successful in predicting human motor behavior than any other type of model. However, the adoption of optimal control theory as a prominent research tool in biomechanics has been consistently hindered by two problems. First, the characteristics of biomechanical simulation models are often not well suited to the formalized solution techniques for optimal control theory. Second, the competencies required to create models and perform simulations are often outside the core training received by researchers in biomechanics and motor control, and the time course for learning these methods and developing valid simulations models from scratch can take years.
The first problem was addressed now 20 years ago with the introduction of Parameter Optimization, which converts optimal control problems to nonlinear programming problems by parameterizing the assumed time-varying neuromuscular control functions. This approach has since been used extensively in simulation studies of human movement. The second problem has been addressed more recently by OpenSim. To generate simulations, OpenSim implements a Computed Muscle Control algorithm, which greatly reduces the computational time required for conventional forward dynamics simulations, but relies on assumptions that (i) resultant joint moments are distributed into individual muscle forces according to a minimization rule, and (ii) a time-varying ground reaction force (GRF) at the the foot-floor interface is known ahead of time. These assumptions may not always be desirable, for example when simulating explosive movements where minimization rules are unlikely to apply, or when performing exploratory work on motor control principles. In these situations, it would be useful to have solutions to both problems of optimal control implementation within a common biomechanical modeling and simulation framework.