Nonlinear Closed-Loop Control of an OpenSim Wrist Model: Tuning Using Genetic Algorithm

Forward dynamics simulations of musculoskeletal systems allow predicting joint movement from neuromuscular excitations. OpenSim models present several nonlinearities that make classic control synthesis and closed-loop movement simulation a challenging task. In previous works, we designed closed-loop controllers, based on linear assumptions, for the identification and design of an \(\mathscr {H}_{\infty }\) controller in series with a reciprocal inhibition architecture, to control an OpenSim wrist biomechanical model. Here, we present an optimized and simplified closed-loop controller, using only the reciprocal inhibition module. Genetic Algorithm optimization of a quadratic cost function was employed to tune the controller parameters. Setpoint control from different initial conditions was tested. The lowest cost controller resulted in rise and settling times for flexion of approximately 68 and 162 ms, respectively. For pronation, 34.3 and 168 ms. Such results, which agree in shape, rise, and settling times with those obtained experimentally on literature, suggests that an appropriate controller structure was selected and tuned. Therefore, a GA-based control design of a reciprocal inhibition architecture seems to be an appropriate approach to handle forward dynamics closed-loop simulations of musculoskeletal systems.